2 * H.26L/H.264/AVC/JVT/14496-10/... decoder
3 * Copyright (c) 2003 Michael Niedermayer <michaelni@gmx.at>
5 * This file is part of FFmpeg.
7 * FFmpeg is free software; you can redistribute it and/or
8 * modify it under the terms of the GNU Lesser General Public
9 * License as published by the Free Software Foundation; either
10 * version 2.1 of the License, or (at your option) any later version.
12 * FFmpeg is distributed in the hope that it will be useful,
13 * but WITHOUT ANY WARRANTY; without even the implied warranty of
14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
15 * Lesser General Public License for more details.
17 * You should have received a copy of the GNU Lesser General Public
18 * License along with FFmpeg; if not, write to the Free Software
19 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
24 * H.264 / AVC / MPEG4 part10 codec.
25 * @author Michael Niedermayer <michaelni@gmx.at>
28 #define UNCHECKED_BITSTREAM_READER 1
30 #include "libavutil/imgutils.h"
31 #include "libavutil/opt.h"
34 #include "cabac_functions.h"
37 #include "mpegvideo.h"
40 #include "h264_mvpred.h"
43 #include "rectangle.h"
45 #include "vdpau_internal.h"
46 #include "libavutil/avassert.h"
51 const uint16_t ff_h264_mb_sizes[4] = { 256, 384, 512, 768 };
53 static const uint8_t rem6[QP_MAX_NUM + 1] = {
54 0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2,
55 3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5,
56 0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2,
57 3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5,
61 static const uint8_t div6[QP_MAX_NUM + 1] = {
62 0, 0, 0, 0, 0, 0, 1, 1, 1, 1, 1, 1, 2, 2, 2, 2, 2, 2, 3, 3, 3,
63 3, 3, 3, 4, 4, 4, 4, 4, 4, 5, 5, 5, 5, 5, 5, 6, 6, 6, 6, 6, 6,
64 7, 7, 7, 7, 7, 7, 8, 8, 8, 8, 8, 8, 9, 9, 9, 9, 9, 9, 10, 10, 10,
65 10,10,10,11,11,11,11,11,11,12,12,12,12,12,12,13,13,13, 13, 13, 13,
69 static const enum AVPixelFormat hwaccel_pixfmt_list_h264_jpeg_420[] = {
77 int avpriv_h264_has_num_reorder_frames(AVCodecContext *avctx)
79 H264Context *h = avctx->priv_data;
80 return h ? h->sps.num_reorder_frames : 0;
84 * Check if the top & left blocks are available if needed and
85 * change the dc mode so it only uses the available blocks.
87 int ff_h264_check_intra4x4_pred_mode(H264Context *h)
89 MpegEncContext *const s = &h->s;
90 static const int8_t top[12] = {
91 -1, 0, LEFT_DC_PRED, -1, -1, -1, -1, -1, 0
93 static const int8_t left[12] = {
94 0, -1, TOP_DC_PRED, 0, -1, -1, -1, 0, -1, DC_128_PRED
98 if (!(h->top_samples_available & 0x8000)) {
99 for (i = 0; i < 4; i++) {
100 int status = top[h->intra4x4_pred_mode_cache[scan8[0] + i]];
102 av_log(h->s.avctx, AV_LOG_ERROR,
103 "top block unavailable for requested intra4x4 mode %d at %d %d\n",
104 status, s->mb_x, s->mb_y);
107 h->intra4x4_pred_mode_cache[scan8[0] + i] = status;
112 if ((h->left_samples_available & 0x8888) != 0x8888) {
113 static const int mask[4] = { 0x8000, 0x2000, 0x80, 0x20 };
114 for (i = 0; i < 4; i++)
115 if (!(h->left_samples_available & mask[i])) {
116 int status = left[h->intra4x4_pred_mode_cache[scan8[0] + 8 * i]];
118 av_log(h->s.avctx, AV_LOG_ERROR,
119 "left block unavailable for requested intra4x4 mode %d at %d %d\n",
120 status, s->mb_x, s->mb_y);
123 h->intra4x4_pred_mode_cache[scan8[0] + 8 * i] = status;
129 } // FIXME cleanup like ff_h264_check_intra_pred_mode
132 * Check if the top & left blocks are available if needed and
133 * change the dc mode so it only uses the available blocks.
135 int ff_h264_check_intra_pred_mode(H264Context *h, int mode, int is_chroma)
137 MpegEncContext *const s = &h->s;
138 static const int8_t top[7] = { LEFT_DC_PRED8x8, 1, -1, -1 };
139 static const int8_t left[7] = { TOP_DC_PRED8x8, -1, 2, -1, DC_128_PRED8x8 };
142 av_log(h->s.avctx, AV_LOG_ERROR,
143 "out of range intra chroma pred mode at %d %d\n",
148 if (!(h->top_samples_available & 0x8000)) {
151 av_log(h->s.avctx, AV_LOG_ERROR,
152 "top block unavailable for requested intra mode at %d %d\n",
158 if ((h->left_samples_available & 0x8080) != 0x8080) {
160 if (is_chroma && (h->left_samples_available & 0x8080)) {
161 // mad cow disease mode, aka MBAFF + constrained_intra_pred
162 mode = ALZHEIMER_DC_L0T_PRED8x8 +
163 (!(h->left_samples_available & 0x8000)) +
164 2 * (mode == DC_128_PRED8x8);
167 av_log(h->s.avctx, AV_LOG_ERROR,
168 "left block unavailable for requested intra mode at %d %d\n",
177 const uint8_t *ff_h264_decode_nal(H264Context *h, const uint8_t *src,
178 int *dst_length, int *consumed, int length)
184 // src[0]&0x80; // forbidden bit
185 h->nal_ref_idc = src[0] >> 5;
186 h->nal_unit_type = src[0] & 0x1F;
191 #define STARTCODE_TEST \
192 if (i + 2 < length && src[i + 1] == 0 && src[i + 2] <= 3) { \
193 if (src[i + 2] != 3) { \
194 /* startcode, so we must be past the end */ \
199 #if HAVE_FAST_UNALIGNED
200 #define FIND_FIRST_ZERO \
201 if (i > 0 && !src[i]) \
206 for (i = 0; i + 1 < length; i += 9) {
207 if (!((~AV_RN64A(src + i) &
208 (AV_RN64A(src + i) - 0x0100010001000101ULL)) &
209 0x8000800080008080ULL))
216 for (i = 0; i + 1 < length; i += 5) {
217 if (!((~AV_RN32A(src + i) &
218 (AV_RN32A(src + i) - 0x01000101U)) &
227 for (i = 0; i + 1 < length; i += 2) {
230 if (i > 0 && src[i - 1] == 0)
236 // use second escape buffer for inter data
237 bufidx = h->nal_unit_type == NAL_DPC ? 1 : 0;
239 si = h->rbsp_buffer_size[bufidx];
240 av_fast_padded_malloc(&h->rbsp_buffer[bufidx], &h->rbsp_buffer_size[bufidx], length+MAX_MBPAIR_SIZE);
241 dst = h->rbsp_buffer[bufidx];
246 if(i>=length-1){ //no escaped 0
248 *consumed= length+1; //+1 for the header
249 if(h->s.avctx->flags2 & CODEC_FLAG2_FAST){
252 memcpy(dst, src, length);
259 while (si + 2 < length) {
260 // remove escapes (very rare 1:2^22)
261 if (src[si + 2] > 3) {
262 dst[di++] = src[si++];
263 dst[di++] = src[si++];
264 } else if (src[si] == 0 && src[si + 1] == 0) {
265 if (src[si + 2] == 3) { // escape
270 } else // next start code
274 dst[di++] = src[si++];
277 dst[di++] = src[si++];
280 memset(dst + di, 0, FF_INPUT_BUFFER_PADDING_SIZE);
283 *consumed = si + 1; // +1 for the header
284 /* FIXME store exact number of bits in the getbitcontext
285 * (it is needed for decoding) */
290 * Identify the exact end of the bitstream
291 * @return the length of the trailing, or 0 if damaged
293 static int decode_rbsp_trailing(H264Context *h, const uint8_t *src)
298 tprintf(h->s.avctx, "rbsp trailing %X\n", v);
300 for (r = 1; r < 9; r++) {
308 static inline int get_lowest_part_list_y(H264Context *h, Picture *pic, int n,
309 int height, int y_offset, int list)
311 int raw_my = h->mv_cache[list][scan8[n]][1];
312 int filter_height = (raw_my & 3) ? 2 : 0;
313 int full_my = (raw_my >> 2) + y_offset;
314 int top = full_my - filter_height;
315 int bottom = full_my + filter_height + height;
317 return FFMAX(abs(top), bottom);
320 static inline void get_lowest_part_y(H264Context *h, int refs[2][48], int n,
321 int height, int y_offset, int list0,
322 int list1, int *nrefs)
324 MpegEncContext *const s = &h->s;
327 y_offset += 16 * (s->mb_y >> MB_FIELD);
330 int ref_n = h->ref_cache[0][scan8[n]];
331 Picture *ref = &h->ref_list[0][ref_n];
333 // Error resilience puts the current picture in the ref list.
334 // Don't try to wait on these as it will cause a deadlock.
335 // Fields can wait on each other, though.
336 if (ref->f.thread_opaque != s->current_picture.f.thread_opaque ||
337 (ref->f.reference & 3) != s->picture_structure) {
338 my = get_lowest_part_list_y(h, ref, n, height, y_offset, 0);
339 if (refs[0][ref_n] < 0)
341 refs[0][ref_n] = FFMAX(refs[0][ref_n], my);
346 int ref_n = h->ref_cache[1][scan8[n]];
347 Picture *ref = &h->ref_list[1][ref_n];
349 if (ref->f.thread_opaque != s->current_picture.f.thread_opaque ||
350 (ref->f.reference & 3) != s->picture_structure) {
351 my = get_lowest_part_list_y(h, ref, n, height, y_offset, 1);
352 if (refs[1][ref_n] < 0)
354 refs[1][ref_n] = FFMAX(refs[1][ref_n], my);
360 * Wait until all reference frames are available for MC operations.
362 * @param h the H264 context
364 static void await_references(H264Context *h)
366 MpegEncContext *const s = &h->s;
367 const int mb_xy = h->mb_xy;
368 const int mb_type = s->current_picture.f.mb_type[mb_xy];
370 int nrefs[2] = { 0 };
373 memset(refs, -1, sizeof(refs));
375 if (IS_16X16(mb_type)) {
376 get_lowest_part_y(h, refs, 0, 16, 0,
377 IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1), nrefs);
378 } else if (IS_16X8(mb_type)) {
379 get_lowest_part_y(h, refs, 0, 8, 0,
380 IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1), nrefs);
381 get_lowest_part_y(h, refs, 8, 8, 8,
382 IS_DIR(mb_type, 1, 0), IS_DIR(mb_type, 1, 1), nrefs);
383 } else if (IS_8X16(mb_type)) {
384 get_lowest_part_y(h, refs, 0, 16, 0,
385 IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1), nrefs);
386 get_lowest_part_y(h, refs, 4, 16, 0,
387 IS_DIR(mb_type, 1, 0), IS_DIR(mb_type, 1, 1), nrefs);
391 av_assert2(IS_8X8(mb_type));
393 for (i = 0; i < 4; i++) {
394 const int sub_mb_type = h->sub_mb_type[i];
396 int y_offset = (i & 2) << 2;
398 if (IS_SUB_8X8(sub_mb_type)) {
399 get_lowest_part_y(h, refs, n, 8, y_offset,
400 IS_DIR(sub_mb_type, 0, 0),
401 IS_DIR(sub_mb_type, 0, 1),
403 } else if (IS_SUB_8X4(sub_mb_type)) {
404 get_lowest_part_y(h, refs, n, 4, y_offset,
405 IS_DIR(sub_mb_type, 0, 0),
406 IS_DIR(sub_mb_type, 0, 1),
408 get_lowest_part_y(h, refs, n + 2, 4, y_offset + 4,
409 IS_DIR(sub_mb_type, 0, 0),
410 IS_DIR(sub_mb_type, 0, 1),
412 } else if (IS_SUB_4X8(sub_mb_type)) {
413 get_lowest_part_y(h, refs, n, 8, y_offset,
414 IS_DIR(sub_mb_type, 0, 0),
415 IS_DIR(sub_mb_type, 0, 1),
417 get_lowest_part_y(h, refs, n + 1, 8, y_offset,
418 IS_DIR(sub_mb_type, 0, 0),
419 IS_DIR(sub_mb_type, 0, 1),
423 av_assert2(IS_SUB_4X4(sub_mb_type));
424 for (j = 0; j < 4; j++) {
425 int sub_y_offset = y_offset + 2 * (j & 2);
426 get_lowest_part_y(h, refs, n + j, 4, sub_y_offset,
427 IS_DIR(sub_mb_type, 0, 0),
428 IS_DIR(sub_mb_type, 0, 1),
435 for (list = h->list_count - 1; list >= 0; list--)
436 for (ref = 0; ref < 48 && nrefs[list]; ref++) {
437 int row = refs[list][ref];
439 Picture *ref_pic = &h->ref_list[list][ref];
440 int ref_field = ref_pic->f.reference - 1;
441 int ref_field_picture = ref_pic->field_picture;
442 int pic_height = 16 * s->mb_height >> ref_field_picture;
447 if (!FIELD_PICTURE && ref_field_picture) { // frame referencing two fields
448 ff_thread_await_progress(&ref_pic->f,
449 FFMIN((row >> 1) - !(row & 1),
452 ff_thread_await_progress(&ref_pic->f,
453 FFMIN((row >> 1), pic_height - 1),
455 } else if (FIELD_PICTURE && !ref_field_picture) { // field referencing one field of a frame
456 ff_thread_await_progress(&ref_pic->f,
457 FFMIN(row * 2 + ref_field,
460 } else if (FIELD_PICTURE) {
461 ff_thread_await_progress(&ref_pic->f,
462 FFMIN(row, pic_height - 1),
465 ff_thread_await_progress(&ref_pic->f,
466 FFMIN(row, pic_height - 1),
473 static av_always_inline void mc_dir_part(H264Context *h, Picture *pic,
474 int n, int square, int height,
476 uint8_t *dest_y, uint8_t *dest_cb,
478 int src_x_offset, int src_y_offset,
479 qpel_mc_func *qpix_op,
480 h264_chroma_mc_func chroma_op,
481 int pixel_shift, int chroma_idc)
483 MpegEncContext *const s = &h->s;
484 const int mx = h->mv_cache[list][scan8[n]][0] + src_x_offset * 8;
485 int my = h->mv_cache[list][scan8[n]][1] + src_y_offset * 8;
486 const int luma_xy = (mx & 3) + ((my & 3) << 2);
487 int offset = ((mx >> 2) << pixel_shift) + (my >> 2) * h->mb_linesize;
488 uint8_t *src_y = pic->f.data[0] + offset;
489 uint8_t *src_cb, *src_cr;
490 int extra_width = h->emu_edge_width;
491 int extra_height = h->emu_edge_height;
493 const int full_mx = mx >> 2;
494 const int full_my = my >> 2;
495 const int pic_width = 16 * s->mb_width;
496 const int pic_height = 16 * s->mb_height >> MB_FIELD;
504 if (full_mx < 0 - extra_width ||
505 full_my < 0 - extra_height ||
506 full_mx + 16 /*FIXME*/ > pic_width + extra_width ||
507 full_my + 16 /*FIXME*/ > pic_height + extra_height) {
508 s->dsp.emulated_edge_mc(s->edge_emu_buffer,
509 src_y - (2 << pixel_shift) - 2 * h->mb_linesize,
511 16 + 5, 16 + 5 /*FIXME*/, full_mx - 2,
512 full_my - 2, pic_width, pic_height);
513 src_y = s->edge_emu_buffer + (2 << pixel_shift) + 2 * h->mb_linesize;
517 qpix_op[luma_xy](dest_y, src_y, h->mb_linesize); // FIXME try variable height perhaps?
519 qpix_op[luma_xy](dest_y + delta, src_y + delta, h->mb_linesize);
521 if (CONFIG_GRAY && s->flags & CODEC_FLAG_GRAY)
524 if (chroma_idc == 3 /* yuv444 */) {
525 src_cb = pic->f.data[1] + offset;
527 s->dsp.emulated_edge_mc(s->edge_emu_buffer,
528 src_cb - (2 << pixel_shift) - 2 * h->mb_linesize,
530 16 + 5, 16 + 5 /*FIXME*/,
531 full_mx - 2, full_my - 2,
532 pic_width, pic_height);
533 src_cb = s->edge_emu_buffer + (2 << pixel_shift) + 2 * h->mb_linesize;
535 qpix_op[luma_xy](dest_cb, src_cb, h->mb_linesize); // FIXME try variable height perhaps?
537 qpix_op[luma_xy](dest_cb + delta, src_cb + delta, h->mb_linesize);
539 src_cr = pic->f.data[2] + offset;
541 s->dsp.emulated_edge_mc(s->edge_emu_buffer,
542 src_cr - (2 << pixel_shift) - 2 * h->mb_linesize,
544 16 + 5, 16 + 5 /*FIXME*/,
545 full_mx - 2, full_my - 2,
546 pic_width, pic_height);
547 src_cr = s->edge_emu_buffer + (2 << pixel_shift) + 2 * h->mb_linesize;
549 qpix_op[luma_xy](dest_cr, src_cr, h->mb_linesize); // FIXME try variable height perhaps?
551 qpix_op[luma_xy](dest_cr + delta, src_cr + delta, h->mb_linesize);
555 ysh = 3 - (chroma_idc == 2 /* yuv422 */);
556 if (chroma_idc == 1 /* yuv420 */ && MB_FIELD) {
557 // chroma offset when predicting from a field of opposite parity
558 my += 2 * ((s->mb_y & 1) - (pic->f.reference - 1));
559 emu |= (my >> 3) < 0 || (my >> 3) + 8 >= (pic_height >> 1);
562 src_cb = pic->f.data[1] + ((mx >> 3) << pixel_shift) +
563 (my >> ysh) * h->mb_uvlinesize;
564 src_cr = pic->f.data[2] + ((mx >> 3) << pixel_shift) +
565 (my >> ysh) * h->mb_uvlinesize;
568 s->dsp.emulated_edge_mc(s->edge_emu_buffer, src_cb, h->mb_uvlinesize,
569 9, 8 * chroma_idc + 1, (mx >> 3), (my >> ysh),
570 pic_width >> 1, pic_height >> (chroma_idc == 1 /* yuv420 */));
571 src_cb = s->edge_emu_buffer;
573 chroma_op(dest_cb, src_cb, h->mb_uvlinesize,
574 height >> (chroma_idc == 1 /* yuv420 */),
575 mx & 7, (my << (chroma_idc == 2 /* yuv422 */)) & 7);
578 s->dsp.emulated_edge_mc(s->edge_emu_buffer, src_cr, h->mb_uvlinesize,
579 9, 8 * chroma_idc + 1, (mx >> 3), (my >> ysh),
580 pic_width >> 1, pic_height >> (chroma_idc == 1 /* yuv420 */));
581 src_cr = s->edge_emu_buffer;
583 chroma_op(dest_cr, src_cr, h->mb_uvlinesize, height >> (chroma_idc == 1 /* yuv420 */),
584 mx & 7, (my << (chroma_idc == 2 /* yuv422 */)) & 7);
587 static av_always_inline void mc_part_std(H264Context *h, int n, int square,
588 int height, int delta,
589 uint8_t *dest_y, uint8_t *dest_cb,
591 int x_offset, int y_offset,
592 qpel_mc_func *qpix_put,
593 h264_chroma_mc_func chroma_put,
594 qpel_mc_func *qpix_avg,
595 h264_chroma_mc_func chroma_avg,
596 int list0, int list1,
597 int pixel_shift, int chroma_idc)
599 MpegEncContext *const s = &h->s;
600 qpel_mc_func *qpix_op = qpix_put;
601 h264_chroma_mc_func chroma_op = chroma_put;
603 dest_y += (2 * x_offset << pixel_shift) + 2 * y_offset * h->mb_linesize;
604 if (chroma_idc == 3 /* yuv444 */) {
605 dest_cb += (2 * x_offset << pixel_shift) + 2 * y_offset * h->mb_linesize;
606 dest_cr += (2 * x_offset << pixel_shift) + 2 * y_offset * h->mb_linesize;
607 } else if (chroma_idc == 2 /* yuv422 */) {
608 dest_cb += (x_offset << pixel_shift) + 2 * y_offset * h->mb_uvlinesize;
609 dest_cr += (x_offset << pixel_shift) + 2 * y_offset * h->mb_uvlinesize;
610 } else { /* yuv420 */
611 dest_cb += (x_offset << pixel_shift) + y_offset * h->mb_uvlinesize;
612 dest_cr += (x_offset << pixel_shift) + y_offset * h->mb_uvlinesize;
614 x_offset += 8 * s->mb_x;
615 y_offset += 8 * (s->mb_y >> MB_FIELD);
618 Picture *ref = &h->ref_list[0][h->ref_cache[0][scan8[n]]];
619 mc_dir_part(h, ref, n, square, height, delta, 0,
620 dest_y, dest_cb, dest_cr, x_offset, y_offset,
621 qpix_op, chroma_op, pixel_shift, chroma_idc);
624 chroma_op = chroma_avg;
628 Picture *ref = &h->ref_list[1][h->ref_cache[1][scan8[n]]];
629 mc_dir_part(h, ref, n, square, height, delta, 1,
630 dest_y, dest_cb, dest_cr, x_offset, y_offset,
631 qpix_op, chroma_op, pixel_shift, chroma_idc);
635 static av_always_inline void mc_part_weighted(H264Context *h, int n, int square,
636 int height, int delta,
637 uint8_t *dest_y, uint8_t *dest_cb,
639 int x_offset, int y_offset,
640 qpel_mc_func *qpix_put,
641 h264_chroma_mc_func chroma_put,
642 h264_weight_func luma_weight_op,
643 h264_weight_func chroma_weight_op,
644 h264_biweight_func luma_weight_avg,
645 h264_biweight_func chroma_weight_avg,
646 int list0, int list1,
647 int pixel_shift, int chroma_idc)
649 MpegEncContext *const s = &h->s;
652 dest_y += (2 * x_offset << pixel_shift) + 2 * y_offset * h->mb_linesize;
653 if (chroma_idc == 3 /* yuv444 */) {
654 chroma_height = height;
655 chroma_weight_avg = luma_weight_avg;
656 chroma_weight_op = luma_weight_op;
657 dest_cb += (2 * x_offset << pixel_shift) + 2 * y_offset * h->mb_linesize;
658 dest_cr += (2 * x_offset << pixel_shift) + 2 * y_offset * h->mb_linesize;
659 } else if (chroma_idc == 2 /* yuv422 */) {
660 chroma_height = height;
661 dest_cb += (x_offset << pixel_shift) + 2 * y_offset * h->mb_uvlinesize;
662 dest_cr += (x_offset << pixel_shift) + 2 * y_offset * h->mb_uvlinesize;
663 } else { /* yuv420 */
664 chroma_height = height >> 1;
665 dest_cb += (x_offset << pixel_shift) + y_offset * h->mb_uvlinesize;
666 dest_cr += (x_offset << pixel_shift) + y_offset * h->mb_uvlinesize;
668 x_offset += 8 * s->mb_x;
669 y_offset += 8 * (s->mb_y >> MB_FIELD);
671 if (list0 && list1) {
672 /* don't optimize for luma-only case, since B-frames usually
673 * use implicit weights => chroma too. */
674 uint8_t *tmp_cb = s->obmc_scratchpad;
675 uint8_t *tmp_cr = s->obmc_scratchpad + (16 << pixel_shift);
676 uint8_t *tmp_y = s->obmc_scratchpad + 16 * h->mb_uvlinesize;
677 int refn0 = h->ref_cache[0][scan8[n]];
678 int refn1 = h->ref_cache[1][scan8[n]];
680 mc_dir_part(h, &h->ref_list[0][refn0], n, square, height, delta, 0,
681 dest_y, dest_cb, dest_cr,
682 x_offset, y_offset, qpix_put, chroma_put,
683 pixel_shift, chroma_idc);
684 mc_dir_part(h, &h->ref_list[1][refn1], n, square, height, delta, 1,
685 tmp_y, tmp_cb, tmp_cr,
686 x_offset, y_offset, qpix_put, chroma_put,
687 pixel_shift, chroma_idc);
689 if (h->use_weight == 2) {
690 int weight0 = h->implicit_weight[refn0][refn1][s->mb_y & 1];
691 int weight1 = 64 - weight0;
692 luma_weight_avg(dest_y, tmp_y, h->mb_linesize,
693 height, 5, weight0, weight1, 0);
694 chroma_weight_avg(dest_cb, tmp_cb, h->mb_uvlinesize,
695 chroma_height, 5, weight0, weight1, 0);
696 chroma_weight_avg(dest_cr, tmp_cr, h->mb_uvlinesize,
697 chroma_height, 5, weight0, weight1, 0);
699 luma_weight_avg(dest_y, tmp_y, h->mb_linesize, height,
700 h->luma_log2_weight_denom,
701 h->luma_weight[refn0][0][0],
702 h->luma_weight[refn1][1][0],
703 h->luma_weight[refn0][0][1] +
704 h->luma_weight[refn1][1][1]);
705 chroma_weight_avg(dest_cb, tmp_cb, h->mb_uvlinesize, chroma_height,
706 h->chroma_log2_weight_denom,
707 h->chroma_weight[refn0][0][0][0],
708 h->chroma_weight[refn1][1][0][0],
709 h->chroma_weight[refn0][0][0][1] +
710 h->chroma_weight[refn1][1][0][1]);
711 chroma_weight_avg(dest_cr, tmp_cr, h->mb_uvlinesize, chroma_height,
712 h->chroma_log2_weight_denom,
713 h->chroma_weight[refn0][0][1][0],
714 h->chroma_weight[refn1][1][1][0],
715 h->chroma_weight[refn0][0][1][1] +
716 h->chroma_weight[refn1][1][1][1]);
719 int list = list1 ? 1 : 0;
720 int refn = h->ref_cache[list][scan8[n]];
721 Picture *ref = &h->ref_list[list][refn];
722 mc_dir_part(h, ref, n, square, height, delta, list,
723 dest_y, dest_cb, dest_cr, x_offset, y_offset,
724 qpix_put, chroma_put, pixel_shift, chroma_idc);
726 luma_weight_op(dest_y, h->mb_linesize, height,
727 h->luma_log2_weight_denom,
728 h->luma_weight[refn][list][0],
729 h->luma_weight[refn][list][1]);
730 if (h->use_weight_chroma) {
731 chroma_weight_op(dest_cb, h->mb_uvlinesize, chroma_height,
732 h->chroma_log2_weight_denom,
733 h->chroma_weight[refn][list][0][0],
734 h->chroma_weight[refn][list][0][1]);
735 chroma_weight_op(dest_cr, h->mb_uvlinesize, chroma_height,
736 h->chroma_log2_weight_denom,
737 h->chroma_weight[refn][list][1][0],
738 h->chroma_weight[refn][list][1][1]);
743 static av_always_inline void prefetch_motion(H264Context *h, int list,
744 int pixel_shift, int chroma_idc)
746 /* fetch pixels for estimated mv 4 macroblocks ahead
747 * optimized for 64byte cache lines */
748 MpegEncContext *const s = &h->s;
749 const int refn = h->ref_cache[list][scan8[0]];
751 const int mx = (h->mv_cache[list][scan8[0]][0] >> 2) + 16 * s->mb_x + 8;
752 const int my = (h->mv_cache[list][scan8[0]][1] >> 2) + 16 * s->mb_y;
753 uint8_t **src = h->ref_list[list][refn].f.data;
754 int off = (mx << pixel_shift) +
755 (my + (s->mb_x & 3) * 4) * h->mb_linesize +
757 s->dsp.prefetch(src[0] + off, s->linesize, 4);
758 if (chroma_idc == 3 /* yuv444 */) {
759 s->dsp.prefetch(src[1] + off, s->linesize, 4);
760 s->dsp.prefetch(src[2] + off, s->linesize, 4);
762 off= (((mx>>1)+64)<<pixel_shift) + ((my>>1) + (s->mb_x&7))*s->uvlinesize;
763 s->dsp.prefetch(src[1] + off, src[2] - src[1], 2);
768 static void free_tables(H264Context *h, int free_rbsp)
773 av_freep(&h->intra4x4_pred_mode);
774 av_freep(&h->chroma_pred_mode_table);
775 av_freep(&h->cbp_table);
776 av_freep(&h->mvd_table[0]);
777 av_freep(&h->mvd_table[1]);
778 av_freep(&h->direct_table);
779 av_freep(&h->non_zero_count);
780 av_freep(&h->slice_table_base);
781 h->slice_table = NULL;
782 av_freep(&h->list_counts);
784 av_freep(&h->mb2b_xy);
785 av_freep(&h->mb2br_xy);
787 for (i = 0; i < MAX_THREADS; i++) {
788 hx = h->thread_context[i];
791 av_freep(&hx->top_borders[1]);
792 av_freep(&hx->top_borders[0]);
793 av_freep(&hx->s.obmc_scratchpad);
795 av_freep(&hx->rbsp_buffer[1]);
796 av_freep(&hx->rbsp_buffer[0]);
797 hx->rbsp_buffer_size[0] = 0;
798 hx->rbsp_buffer_size[1] = 0;
801 av_freep(&h->thread_context[i]);
805 static void init_dequant8_coeff_table(H264Context *h)
808 const int max_qp = 51 + 6 * (h->sps.bit_depth_luma - 8);
810 for (i = 0; i < 6; i++) {
811 h->dequant8_coeff[i] = h->dequant8_buffer[i];
812 for (j = 0; j < i; j++)
813 if (!memcmp(h->pps.scaling_matrix8[j], h->pps.scaling_matrix8[i],
814 64 * sizeof(uint8_t))) {
815 h->dequant8_coeff[i] = h->dequant8_buffer[j];
821 for (q = 0; q < max_qp + 1; q++) {
824 for (x = 0; x < 64; x++)
825 h->dequant8_coeff[i][q][(x >> 3) | ((x & 7) << 3)] =
826 ((uint32_t)dequant8_coeff_init[idx][dequant8_coeff_init_scan[((x >> 1) & 12) | (x & 3)]] *
827 h->pps.scaling_matrix8[i][x]) << shift;
832 static void init_dequant4_coeff_table(H264Context *h)
835 const int max_qp = 51 + 6 * (h->sps.bit_depth_luma - 8);
836 for (i = 0; i < 6; i++) {
837 h->dequant4_coeff[i] = h->dequant4_buffer[i];
838 for (j = 0; j < i; j++)
839 if (!memcmp(h->pps.scaling_matrix4[j], h->pps.scaling_matrix4[i],
840 16 * sizeof(uint8_t))) {
841 h->dequant4_coeff[i] = h->dequant4_buffer[j];
847 for (q = 0; q < max_qp + 1; q++) {
848 int shift = div6[q] + 2;
850 for (x = 0; x < 16; x++)
851 h->dequant4_coeff[i][q][(x >> 2) | ((x << 2) & 0xF)] =
852 ((uint32_t)dequant4_coeff_init[idx][(x & 1) + ((x >> 2) & 1)] *
853 h->pps.scaling_matrix4[i][x]) << shift;
858 static void init_dequant_tables(H264Context *h)
861 init_dequant4_coeff_table(h);
862 if (h->pps.transform_8x8_mode)
863 init_dequant8_coeff_table(h);
864 if (h->sps.transform_bypass) {
865 for (i = 0; i < 6; i++)
866 for (x = 0; x < 16; x++)
867 h->dequant4_coeff[i][0][x] = 1 << 6;
868 if (h->pps.transform_8x8_mode)
869 for (i = 0; i < 6; i++)
870 for (x = 0; x < 64; x++)
871 h->dequant8_coeff[i][0][x] = 1 << 6;
875 int ff_h264_alloc_tables(H264Context *h)
877 MpegEncContext *const s = &h->s;
878 const int big_mb_num = s->mb_stride * (s->mb_height + 1);
879 const int row_mb_num = 2*s->mb_stride*FFMAX(s->avctx->thread_count, 1);
882 FF_ALLOCZ_OR_GOTO(h->s.avctx, h->intra4x4_pred_mode,
883 row_mb_num * 8 * sizeof(uint8_t), fail)
884 FF_ALLOCZ_OR_GOTO(h->s.avctx, h->non_zero_count,
885 big_mb_num * 48 * sizeof(uint8_t), fail)
886 FF_ALLOCZ_OR_GOTO(h->s.avctx, h->slice_table_base,
887 (big_mb_num + s->mb_stride) * sizeof(*h->slice_table_base), fail)
888 FF_ALLOCZ_OR_GOTO(h->s.avctx, h->cbp_table,
889 big_mb_num * sizeof(uint16_t), fail)
890 FF_ALLOCZ_OR_GOTO(h->s.avctx, h->chroma_pred_mode_table,
891 big_mb_num * sizeof(uint8_t), fail)
892 FF_ALLOCZ_OR_GOTO(h->s.avctx, h->mvd_table[0],
893 16 * row_mb_num * sizeof(uint8_t), fail);
894 FF_ALLOCZ_OR_GOTO(h->s.avctx, h->mvd_table[1],
895 16 * row_mb_num * sizeof(uint8_t), fail);
896 FF_ALLOCZ_OR_GOTO(h->s.avctx, h->direct_table,
897 4 * big_mb_num * sizeof(uint8_t), fail);
898 FF_ALLOCZ_OR_GOTO(h->s.avctx, h->list_counts,
899 big_mb_num * sizeof(uint8_t), fail)
901 memset(h->slice_table_base, -1,
902 (big_mb_num + s->mb_stride) * sizeof(*h->slice_table_base));
903 h->slice_table = h->slice_table_base + s->mb_stride * 2 + 1;
905 FF_ALLOCZ_OR_GOTO(h->s.avctx, h->mb2b_xy,
906 big_mb_num * sizeof(uint32_t), fail);
907 FF_ALLOCZ_OR_GOTO(h->s.avctx, h->mb2br_xy,
908 big_mb_num * sizeof(uint32_t), fail);
909 for (y = 0; y < s->mb_height; y++)
910 for (x = 0; x < s->mb_width; x++) {
911 const int mb_xy = x + y * s->mb_stride;
912 const int b_xy = 4 * x + 4 * y * h->b_stride;
914 h->mb2b_xy[mb_xy] = b_xy;
915 h->mb2br_xy[mb_xy] = 8 * (FMO ? mb_xy : (mb_xy % (2 * s->mb_stride)));
918 s->obmc_scratchpad = NULL;
920 if (!h->dequant4_coeff[0])
921 init_dequant_tables(h);
931 * Mimic alloc_tables(), but for every context thread.
933 static void clone_tables(H264Context *dst, H264Context *src, int i)
935 MpegEncContext *const s = &src->s;
936 dst->intra4x4_pred_mode = src->intra4x4_pred_mode + i * 8 * 2 * s->mb_stride;
937 dst->non_zero_count = src->non_zero_count;
938 dst->slice_table = src->slice_table;
939 dst->cbp_table = src->cbp_table;
940 dst->mb2b_xy = src->mb2b_xy;
941 dst->mb2br_xy = src->mb2br_xy;
942 dst->chroma_pred_mode_table = src->chroma_pred_mode_table;
943 dst->mvd_table[0] = src->mvd_table[0] + i * 8 * 2 * s->mb_stride;
944 dst->mvd_table[1] = src->mvd_table[1] + i * 8 * 2 * s->mb_stride;
945 dst->direct_table = src->direct_table;
946 dst->list_counts = src->list_counts;
947 dst->s.obmc_scratchpad = NULL;
948 ff_h264_pred_init(&dst->hpc, src->s.codec_id, src->sps.bit_depth_luma,
949 src->sps.chroma_format_idc);
954 * Allocate buffers which are not shared amongst multiple threads.
956 static int context_init(H264Context *h)
958 FF_ALLOCZ_OR_GOTO(h->s.avctx, h->top_borders[0],
959 h->s.mb_width * 16 * 3 * sizeof(uint8_t) * 2, fail)
960 FF_ALLOCZ_OR_GOTO(h->s.avctx, h->top_borders[1],
961 h->s.mb_width * 16 * 3 * sizeof(uint8_t) * 2, fail)
963 h->ref_cache[0][scan8[5] + 1] =
964 h->ref_cache[0][scan8[7] + 1] =
965 h->ref_cache[0][scan8[13] + 1] =
966 h->ref_cache[1][scan8[5] + 1] =
967 h->ref_cache[1][scan8[7] + 1] =
968 h->ref_cache[1][scan8[13] + 1] = PART_NOT_AVAILABLE;
973 return -1; // free_tables will clean up for us
976 static int decode_nal_units(H264Context *h, const uint8_t *buf, int buf_size);
978 static av_cold void common_init(H264Context *h)
980 MpegEncContext *const s = &h->s;
982 s->width = s->avctx->width;
983 s->height = s->avctx->height;
984 s->codec_id = s->avctx->codec->id;
986 s->avctx->bits_per_raw_sample = 8;
987 h->cur_chroma_format_idc = 1;
989 ff_h264dsp_init(&h->h264dsp,
990 s->avctx->bits_per_raw_sample, h->cur_chroma_format_idc);
991 ff_h264_pred_init(&h->hpc, s->codec_id,
992 s->avctx->bits_per_raw_sample, h->cur_chroma_format_idc);
994 h->dequant_coeff_pps = -1;
995 s->unrestricted_mv = 1;
997 s->dsp.dct_bits = 16;
998 /* needed so that IDCT permutation is known early */
999 ff_dsputil_init(&s->dsp, s->avctx);
1001 memset(h->pps.scaling_matrix4, 16, 6 * 16 * sizeof(uint8_t));
1002 memset(h->pps.scaling_matrix8, 16, 2 * 64 * sizeof(uint8_t));
1005 static int ff_h264_decode_extradata_internal(H264Context *h, const uint8_t *buf, int size)
1007 AVCodecContext *avctx = h->s.avctx;
1009 if (!buf || size <= 0)
1013 int i, cnt, nalsize;
1014 const unsigned char *p = buf;
1019 av_log(avctx, AV_LOG_ERROR, "avcC too short\n");
1022 /* sps and pps in the avcC always have length coded with 2 bytes,
1023 * so put a fake nal_length_size = 2 while parsing them */
1024 h->nal_length_size = 2;
1025 // Decode sps from avcC
1026 cnt = *(p + 5) & 0x1f; // Number of sps
1028 for (i = 0; i < cnt; i++) {
1029 nalsize = AV_RB16(p) + 2;
1030 if(nalsize > size - (p-buf))
1032 if (decode_nal_units(h, p, nalsize) < 0) {
1033 av_log(avctx, AV_LOG_ERROR,
1034 "Decoding sps %d from avcC failed\n", i);
1039 // Decode pps from avcC
1040 cnt = *(p++); // Number of pps
1041 for (i = 0; i < cnt; i++) {
1042 nalsize = AV_RB16(p) + 2;
1043 if(nalsize > size - (p-buf))
1045 if (decode_nal_units(h, p, nalsize) < 0) {
1046 av_log(avctx, AV_LOG_ERROR,
1047 "Decoding pps %d from avcC failed\n", i);
1052 // Now store right nal length size, that will be used to parse all other nals
1053 h->nal_length_size = (buf[4] & 0x03) + 1;
1056 if (decode_nal_units(h, buf, size) < 0)
1062 int ff_h264_decode_extradata(H264Context *h, const uint8_t *buf, int size)
1065 h->decoding_extradata = 1;
1066 ret = ff_h264_decode_extradata_internal(h, buf, size);
1067 h->decoding_extradata = 0;
1071 av_cold int ff_h264_decode_init(AVCodecContext *avctx)
1073 H264Context *h = avctx->priv_data;
1074 MpegEncContext *const s = &h->s;
1077 ff_MPV_decode_defaults(s);
1082 s->out_format = FMT_H264;
1083 s->workaround_bugs = avctx->workaround_bugs;
1086 // s->decode_mb = ff_h263_decode_mb;
1087 s->quarter_sample = 1;
1088 if (!avctx->has_b_frames)
1091 avctx->chroma_sample_location = AVCHROMA_LOC_LEFT;
1093 ff_h264_decode_init_vlc();
1096 h->sps.bit_depth_luma = avctx->bits_per_raw_sample = 8;
1098 h->thread_context[0] = h;
1099 h->outputed_poc = h->next_outputed_poc = INT_MIN;
1100 for (i = 0; i < MAX_DELAYED_PIC_COUNT; i++)
1101 h->last_pocs[i] = INT_MIN;
1102 h->prev_poc_msb = 1 << 16;
1103 h->prev_frame_num = -1;
1105 ff_h264_reset_sei(h);
1106 if (avctx->codec_id == AV_CODEC_ID_H264) {
1107 if (avctx->ticks_per_frame == 1) {
1108 if(s->avctx->time_base.den < INT_MAX/2) {
1109 s->avctx->time_base.den *= 2;
1111 s->avctx->time_base.num /= 2;
1113 avctx->ticks_per_frame = 2;
1116 if (avctx->extradata_size > 0 && avctx->extradata &&
1117 ff_h264_decode_extradata(h, avctx->extradata, avctx->extradata_size) < 0) {
1118 ff_h264_free_context(h);
1122 if (h->sps.bitstream_restriction_flag &&
1123 s->avctx->has_b_frames < h->sps.num_reorder_frames) {
1124 s->avctx->has_b_frames = h->sps.num_reorder_frames;
1128 ff_init_cabac_states();
1133 #define IN_RANGE(a, b, size) (((a) >= (b)) && ((a) < ((b) + (size))))
1135 static void copy_picture_range(Picture **to, Picture **from, int count,
1136 MpegEncContext *new_base,
1137 MpegEncContext *old_base)
1141 for (i = 0; i < count; i++) {
1142 assert((IN_RANGE(from[i], old_base, sizeof(*old_base)) ||
1143 IN_RANGE(from[i], old_base->picture,
1144 sizeof(Picture) * old_base->picture_count) ||
1146 to[i] = REBASE_PICTURE(from[i], new_base, old_base);
1150 static void copy_parameter_set(void **to, void **from, int count, int size)
1154 for (i = 0; i < count; i++) {
1155 if (to[i] && !from[i])
1157 else if (from[i] && !to[i])
1158 to[i] = av_malloc(size);
1161 memcpy(to[i], from[i], size);
1165 static int decode_init_thread_copy(AVCodecContext *avctx)
1167 H264Context *h = avctx->priv_data;
1169 if (!avctx->internal->is_copy)
1171 memset(h->sps_buffers, 0, sizeof(h->sps_buffers));
1172 memset(h->pps_buffers, 0, sizeof(h->pps_buffers));
1177 #define copy_fields(to, from, start_field, end_field) \
1178 memcpy(&to->start_field, &from->start_field, \
1179 (char *)&to->end_field - (char *)&to->start_field)
1181 static int decode_update_thread_context(AVCodecContext *dst,
1182 const AVCodecContext *src)
1184 H264Context *h = dst->priv_data, *h1 = src->priv_data;
1185 MpegEncContext *const s = &h->s, *const s1 = &h1->s;
1186 int inited = s->context_initialized, err;
1192 err = ff_mpeg_update_thread_context(dst, src);
1196 // FIXME handle width/height changing
1198 for (i = 0; i < MAX_SPS_COUNT; i++)
1199 av_freep(h->sps_buffers + i);
1201 for (i = 0; i < MAX_PPS_COUNT; i++)
1202 av_freep(h->pps_buffers + i);
1204 // copy all fields after MpegEnc
1205 memcpy(&h->s + 1, &h1->s + 1,
1206 sizeof(H264Context) - sizeof(MpegEncContext));
1207 memset(h->sps_buffers, 0, sizeof(h->sps_buffers));
1208 memset(h->pps_buffers, 0, sizeof(h->pps_buffers));
1210 if (s1->context_initialized) {
1211 if (ff_h264_alloc_tables(h) < 0) {
1212 av_log(dst, AV_LOG_ERROR, "Could not allocate memory for h264\n");
1213 return AVERROR(ENOMEM);
1217 /* frame_start may not be called for the next thread (if it's decoding
1218 * a bottom field) so this has to be allocated here */
1219 h->s.obmc_scratchpad = av_malloc(16 * 6 * s->linesize);
1222 for (i = 0; i < 2; i++) {
1223 h->rbsp_buffer[i] = NULL;
1224 h->rbsp_buffer_size[i] = 0;
1227 h->thread_context[0] = h;
1229 s->dsp.clear_blocks(h->mb);
1230 s->dsp.clear_blocks(h->mb + (24 * 16 << h->pixel_shift));
1233 // extradata/NAL handling
1234 h->is_avc = h1->is_avc;
1237 copy_parameter_set((void **)h->sps_buffers, (void **)h1->sps_buffers,
1238 MAX_SPS_COUNT, sizeof(SPS));
1240 copy_parameter_set((void **)h->pps_buffers, (void **)h1->pps_buffers,
1241 MAX_PPS_COUNT, sizeof(PPS));
1244 // Dequantization matrices
1245 // FIXME these are big - can they be only copied when PPS changes?
1246 copy_fields(h, h1, dequant4_buffer, dequant4_coeff);
1248 for (i = 0; i < 6; i++)
1249 h->dequant4_coeff[i] = h->dequant4_buffer[0] +
1250 (h1->dequant4_coeff[i] - h1->dequant4_buffer[0]);
1252 for (i = 0; i < 6; i++)
1253 h->dequant8_coeff[i] = h->dequant8_buffer[0] +
1254 (h1->dequant8_coeff[i] - h1->dequant8_buffer[0]);
1256 h->dequant_coeff_pps = h1->dequant_coeff_pps;
1259 copy_fields(h, h1, poc_lsb, redundant_pic_count);
1262 copy_fields(h, h1, ref_count, list_count);
1263 copy_fields(h, h1, ref_list, intra_gb);
1264 copy_fields(h, h1, short_ref, cabac_init_idc);
1266 copy_picture_range(h->short_ref, h1->short_ref, 32, s, s1);
1267 copy_picture_range(h->long_ref, h1->long_ref, 32, s, s1);
1268 copy_picture_range(h->delayed_pic, h1->delayed_pic,
1269 MAX_DELAYED_PIC_COUNT + 2, s, s1);
1271 h->last_slice_type = h1->last_slice_type;
1274 if (!s->current_picture_ptr)
1278 err = ff_h264_execute_ref_pic_marking(h, h->mmco, h->mmco_index);
1279 h->prev_poc_msb = h->poc_msb;
1280 h->prev_poc_lsb = h->poc_lsb;
1282 h->prev_frame_num_offset = h->frame_num_offset;
1283 h->prev_frame_num = h->frame_num;
1284 h->outputed_poc = h->next_outputed_poc;
1289 int ff_h264_frame_start(H264Context *h)
1291 MpegEncContext *const s = &h->s;
1293 const int pixel_shift = h->pixel_shift;
1295 if (ff_MPV_frame_start(s, s->avctx) < 0)
1297 ff_er_frame_start(s);
1299 * ff_MPV_frame_start uses pict_type to derive key_frame.
1300 * This is incorrect for H.264; IDR markings must be used.
1301 * Zero here; IDR markings per slice in frame or fields are ORed in later.
1302 * See decode_nal_units().
1304 s->current_picture_ptr->f.key_frame = 0;
1305 s->current_picture_ptr->sync = 0;
1306 s->current_picture_ptr->mmco_reset = 0;
1308 assert(s->linesize && s->uvlinesize);
1310 for (i = 0; i < 16; i++) {
1311 h->block_offset[i] = (4 * ((scan8[i] - scan8[0]) & 7) << pixel_shift) + 4 * s->linesize * ((scan8[i] - scan8[0]) >> 3);
1312 h->block_offset[48 + i] = (4 * ((scan8[i] - scan8[0]) & 7) << pixel_shift) + 8 * s->linesize * ((scan8[i] - scan8[0]) >> 3);
1314 for (i = 0; i < 16; i++) {
1315 h->block_offset[16 + i] =
1316 h->block_offset[32 + i] = (4 * ((scan8[i] - scan8[0]) & 7) << pixel_shift) + 4 * s->uvlinesize * ((scan8[i] - scan8[0]) >> 3);
1317 h->block_offset[48 + 16 + i] =
1318 h->block_offset[48 + 32 + i] = (4 * ((scan8[i] - scan8[0]) & 7) << pixel_shift) + 8 * s->uvlinesize * ((scan8[i] - scan8[0]) >> 3);
1321 /* can't be in alloc_tables because linesize isn't known there.
1322 * FIXME: redo bipred weight to not require extra buffer? */
1323 for (i = 0; i < s->slice_context_count; i++)
1324 if (h->thread_context[i] && !h->thread_context[i]->s.obmc_scratchpad)
1325 h->thread_context[i]->s.obmc_scratchpad = av_malloc(16 * 6 * s->linesize);
1327 /* Some macroblocks can be accessed before they're available in case
1328 * of lost slices, MBAFF or threading. */
1329 memset(h->slice_table, -1,
1330 (s->mb_height * s->mb_stride - 1) * sizeof(*h->slice_table));
1332 // s->decode = (s->flags & CODEC_FLAG_PSNR) || !s->encoding ||
1333 // s->current_picture.f.reference /* || h->contains_intra */ || 1;
1335 /* We mark the current picture as non-reference after allocating it, so
1336 * that if we break out due to an error it can be released automatically
1337 * in the next ff_MPV_frame_start().
1338 * SVQ3 as well as most other codecs have only last/next/current and thus
1339 * get released even with set reference, besides SVQ3 and others do not
1340 * mark frames as reference later "naturally". */
1341 if (s->codec_id != AV_CODEC_ID_SVQ3)
1342 s->current_picture_ptr->f.reference = 0;
1344 s->current_picture_ptr->field_poc[0] =
1345 s->current_picture_ptr->field_poc[1] = INT_MAX;
1347 h->next_output_pic = NULL;
1349 assert(s->current_picture_ptr->long_ref == 0);
1355 * Run setup operations that must be run after slice header decoding.
1356 * This includes finding the next displayed frame.
1358 * @param h h264 master context
1359 * @param setup_finished enough NALs have been read that we can call
1360 * ff_thread_finish_setup()
1362 static void decode_postinit(H264Context *h, int setup_finished)
1364 MpegEncContext *const s = &h->s;
1365 Picture *out = s->current_picture_ptr;
1366 Picture *cur = s->current_picture_ptr;
1367 int i, pics, out_of_order, out_idx;
1369 s->current_picture_ptr->f.qscale_type = FF_QSCALE_TYPE_H264;
1370 s->current_picture_ptr->f.pict_type = s->pict_type;
1372 if (h->next_output_pic)
1375 if (cur->field_poc[0] == INT_MAX || cur->field_poc[1] == INT_MAX) {
1376 /* FIXME: if we have two PAFF fields in one packet, we can't start
1377 * the next thread here. If we have one field per packet, we can.
1378 * The check in decode_nal_units() is not good enough to find this
1379 * yet, so we assume the worst for now. */
1380 // if (setup_finished)
1381 // ff_thread_finish_setup(s->avctx);
1385 cur->f.interlaced_frame = 0;
1386 cur->f.repeat_pict = 0;
1388 /* Signal interlacing information externally. */
1389 /* Prioritize picture timing SEI information over used
1390 * decoding process if it exists. */
1392 if (h->sps.pic_struct_present_flag) {
1393 switch (h->sei_pic_struct) {
1394 case SEI_PIC_STRUCT_FRAME:
1396 case SEI_PIC_STRUCT_TOP_FIELD:
1397 case SEI_PIC_STRUCT_BOTTOM_FIELD:
1398 cur->f.interlaced_frame = 1;
1400 case SEI_PIC_STRUCT_TOP_BOTTOM:
1401 case SEI_PIC_STRUCT_BOTTOM_TOP:
1402 if (FIELD_OR_MBAFF_PICTURE)
1403 cur->f.interlaced_frame = 1;
1405 // try to flag soft telecine progressive
1406 cur->f.interlaced_frame = h->prev_interlaced_frame;
1408 case SEI_PIC_STRUCT_TOP_BOTTOM_TOP:
1409 case SEI_PIC_STRUCT_BOTTOM_TOP_BOTTOM:
1410 /* Signal the possibility of telecined film externally
1411 * (pic_struct 5,6). From these hints, let the applications
1412 * decide if they apply deinterlacing. */
1413 cur->f.repeat_pict = 1;
1415 case SEI_PIC_STRUCT_FRAME_DOUBLING:
1416 // Force progressive here, doubling interlaced frame is a bad idea.
1417 cur->f.repeat_pict = 2;
1419 case SEI_PIC_STRUCT_FRAME_TRIPLING:
1420 cur->f.repeat_pict = 4;
1424 if ((h->sei_ct_type & 3) &&
1425 h->sei_pic_struct <= SEI_PIC_STRUCT_BOTTOM_TOP)
1426 cur->f.interlaced_frame = (h->sei_ct_type & (1 << 1)) != 0;
1428 /* Derive interlacing flag from used decoding process. */
1429 cur->f.interlaced_frame = FIELD_OR_MBAFF_PICTURE;
1431 h->prev_interlaced_frame = cur->f.interlaced_frame;
1433 if (cur->field_poc[0] != cur->field_poc[1]) {
1434 /* Derive top_field_first from field pocs. */
1435 cur->f.top_field_first = cur->field_poc[0] < cur->field_poc[1];
1437 if (cur->f.interlaced_frame || h->sps.pic_struct_present_flag) {
1438 /* Use picture timing SEI information. Even if it is a
1439 * information of a past frame, better than nothing. */
1440 if (h->sei_pic_struct == SEI_PIC_STRUCT_TOP_BOTTOM ||
1441 h->sei_pic_struct == SEI_PIC_STRUCT_TOP_BOTTOM_TOP)
1442 cur->f.top_field_first = 1;
1444 cur->f.top_field_first = 0;
1446 /* Most likely progressive */
1447 cur->f.top_field_first = 0;
1451 cur->mmco_reset = h->mmco_reset;
1453 // FIXME do something with unavailable reference frames
1455 /* Sort B-frames into display order */
1457 if (h->sps.bitstream_restriction_flag &&
1458 s->avctx->has_b_frames < h->sps.num_reorder_frames) {
1459 s->avctx->has_b_frames = h->sps.num_reorder_frames;
1463 if (s->avctx->strict_std_compliance >= FF_COMPLIANCE_STRICT &&
1464 !h->sps.bitstream_restriction_flag) {
1465 s->avctx->has_b_frames = MAX_DELAYED_PIC_COUNT - 1;
1469 for (i = 0; 1; i++) {
1470 if(i == MAX_DELAYED_PIC_COUNT || cur->poc < h->last_pocs[i]){
1472 h->last_pocs[i-1] = cur->poc;
1475 h->last_pocs[i-1]= h->last_pocs[i];
1478 out_of_order = MAX_DELAYED_PIC_COUNT - i;
1479 if( cur->f.pict_type == AV_PICTURE_TYPE_B
1480 || (h->last_pocs[MAX_DELAYED_PIC_COUNT-2] > INT_MIN && h->last_pocs[MAX_DELAYED_PIC_COUNT-1] - h->last_pocs[MAX_DELAYED_PIC_COUNT-2] > 2))
1481 out_of_order = FFMAX(out_of_order, 1);
1482 if(s->avctx->has_b_frames < out_of_order && !h->sps.bitstream_restriction_flag){
1483 av_log(s->avctx, AV_LOG_VERBOSE, "Increasing reorder buffer to %d\n", out_of_order);
1484 s->avctx->has_b_frames = out_of_order;
1489 while (h->delayed_pic[pics])
1492 av_assert0(pics <= MAX_DELAYED_PIC_COUNT);
1494 h->delayed_pic[pics++] = cur;
1495 if (cur->f.reference == 0)
1496 cur->f.reference = DELAYED_PIC_REF;
1498 out = h->delayed_pic[0];
1500 for (i = 1; h->delayed_pic[i] &&
1501 !h->delayed_pic[i]->f.key_frame &&
1502 !h->delayed_pic[i]->mmco_reset;
1504 if (h->delayed_pic[i]->poc < out->poc) {
1505 out = h->delayed_pic[i];
1508 if (s->avctx->has_b_frames == 0 &&
1509 (h->delayed_pic[0]->f.key_frame || h->delayed_pic[0]->mmco_reset))
1510 h->next_outputed_poc = INT_MIN;
1511 out_of_order = out->poc < h->next_outputed_poc;
1513 if (out_of_order || pics > s->avctx->has_b_frames) {
1514 out->f.reference &= ~DELAYED_PIC_REF;
1515 // for frame threading, the owner must be the second field's thread or
1516 // else the first thread can release the picture and reuse it unsafely
1518 for (i = out_idx; h->delayed_pic[i]; i++)
1519 h->delayed_pic[i] = h->delayed_pic[i + 1];
1521 if (!out_of_order && pics > s->avctx->has_b_frames) {
1522 h->next_output_pic = out;
1523 if (out_idx == 0 && h->delayed_pic[0] && (h->delayed_pic[0]->f.key_frame || h->delayed_pic[0]->mmco_reset)) {
1524 h->next_outputed_poc = INT_MIN;
1526 h->next_outputed_poc = out->poc;
1528 av_log(s->avctx, AV_LOG_DEBUG, "no picture %s\n", out_of_order ? "ooo" : "");
1531 if (h->next_output_pic && h->next_output_pic->sync) {
1536 ff_thread_finish_setup(s->avctx);
1539 static av_always_inline void backup_mb_border(H264Context *h, uint8_t *src_y,
1540 uint8_t *src_cb, uint8_t *src_cr,
1541 int linesize, int uvlinesize,
1544 MpegEncContext *const s = &h->s;
1545 uint8_t *top_border;
1547 const int pixel_shift = h->pixel_shift;
1548 int chroma444 = CHROMA444;
1549 int chroma422 = CHROMA422;
1552 src_cb -= uvlinesize;
1553 src_cr -= uvlinesize;
1555 if (!simple && FRAME_MBAFF) {
1558 top_border = h->top_borders[0][s->mb_x];
1559 AV_COPY128(top_border, src_y + 15 * linesize);
1561 AV_COPY128(top_border + 16, src_y + 15 * linesize + 16);
1562 if (simple || !CONFIG_GRAY || !(s->flags & CODEC_FLAG_GRAY)) {
1565 AV_COPY128(top_border + 32, src_cb + 15 * uvlinesize);
1566 AV_COPY128(top_border + 48, src_cb + 15 * uvlinesize + 16);
1567 AV_COPY128(top_border + 64, src_cr + 15 * uvlinesize);
1568 AV_COPY128(top_border + 80, src_cr + 15 * uvlinesize + 16);
1570 AV_COPY128(top_border + 16, src_cb + 15 * uvlinesize);
1571 AV_COPY128(top_border + 32, src_cr + 15 * uvlinesize);
1573 } else if (chroma422) {
1575 AV_COPY128(top_border + 32, src_cb + 15 * uvlinesize);
1576 AV_COPY128(top_border + 48, src_cr + 15 * uvlinesize);
1578 AV_COPY64(top_border + 16, src_cb + 15 * uvlinesize);
1579 AV_COPY64(top_border + 24, src_cr + 15 * uvlinesize);
1583 AV_COPY128(top_border + 32, src_cb + 7 * uvlinesize);
1584 AV_COPY128(top_border + 48, src_cr + 7 * uvlinesize);
1586 AV_COPY64(top_border + 16, src_cb + 7 * uvlinesize);
1587 AV_COPY64(top_border + 24, src_cr + 7 * uvlinesize);
1592 } else if (MB_MBAFF) {
1598 top_border = h->top_borders[top_idx][s->mb_x];
1599 /* There are two lines saved, the line above the top macroblock
1600 * of a pair, and the line above the bottom macroblock. */
1601 AV_COPY128(top_border, src_y + 16 * linesize);
1603 AV_COPY128(top_border + 16, src_y + 16 * linesize + 16);
1605 if (simple || !CONFIG_GRAY || !(s->flags & CODEC_FLAG_GRAY)) {
1608 AV_COPY128(top_border + 32, src_cb + 16 * linesize);
1609 AV_COPY128(top_border + 48, src_cb + 16 * linesize + 16);
1610 AV_COPY128(top_border + 64, src_cr + 16 * linesize);
1611 AV_COPY128(top_border + 80, src_cr + 16 * linesize + 16);
1613 AV_COPY128(top_border + 16, src_cb + 16 * linesize);
1614 AV_COPY128(top_border + 32, src_cr + 16 * linesize);
1616 } else if (chroma422) {
1618 AV_COPY128(top_border + 32, src_cb + 16 * uvlinesize);
1619 AV_COPY128(top_border + 48, src_cr + 16 * uvlinesize);
1621 AV_COPY64(top_border + 16, src_cb + 16 * uvlinesize);
1622 AV_COPY64(top_border + 24, src_cr + 16 * uvlinesize);
1626 AV_COPY128(top_border + 32, src_cb + 8 * uvlinesize);
1627 AV_COPY128(top_border + 48, src_cr + 8 * uvlinesize);
1629 AV_COPY64(top_border + 16, src_cb + 8 * uvlinesize);
1630 AV_COPY64(top_border + 24, src_cr + 8 * uvlinesize);
1636 static av_always_inline void xchg_mb_border(H264Context *h, uint8_t *src_y,
1637 uint8_t *src_cb, uint8_t *src_cr,
1638 int linesize, int uvlinesize,
1639 int xchg, int chroma444,
1640 int simple, int pixel_shift)
1642 MpegEncContext *const s = &h->s;
1643 int deblock_topleft;
1646 uint8_t *top_border_m1;
1647 uint8_t *top_border;
1649 if (!simple && FRAME_MBAFF) {
1654 top_idx = MB_MBAFF ? 0 : 1;
1658 if (h->deblocking_filter == 2) {
1659 deblock_topleft = h->slice_table[h->mb_xy - 1 - s->mb_stride] == h->slice_num;
1660 deblock_top = h->top_type;
1662 deblock_topleft = (s->mb_x > 0);
1663 deblock_top = (s->mb_y > !!MB_FIELD);
1666 src_y -= linesize + 1 + pixel_shift;
1667 src_cb -= uvlinesize + 1 + pixel_shift;
1668 src_cr -= uvlinesize + 1 + pixel_shift;
1670 top_border_m1 = h->top_borders[top_idx][s->mb_x - 1];
1671 top_border = h->top_borders[top_idx][s->mb_x];
1673 #define XCHG(a, b, xchg) \
1674 if (pixel_shift) { \
1676 AV_SWAP64(b + 0, a + 0); \
1677 AV_SWAP64(b + 8, a + 8); \
1687 if (deblock_topleft) {
1688 XCHG(top_border_m1 + (8 << pixel_shift),
1689 src_y - (7 << pixel_shift), 1);
1691 XCHG(top_border + (0 << pixel_shift), src_y + (1 << pixel_shift), xchg);
1692 XCHG(top_border + (8 << pixel_shift), src_y + (9 << pixel_shift), 1);
1693 if (s->mb_x + 1 < s->mb_width) {
1694 XCHG(h->top_borders[top_idx][s->mb_x + 1],
1695 src_y + (17 << pixel_shift), 1);
1698 if (simple || !CONFIG_GRAY || !(s->flags & CODEC_FLAG_GRAY)) {
1700 if (deblock_topleft) {
1701 XCHG(top_border_m1 + (24 << pixel_shift), src_cb - (7 << pixel_shift), 1);
1702 XCHG(top_border_m1 + (40 << pixel_shift), src_cr - (7 << pixel_shift), 1);
1704 XCHG(top_border + (16 << pixel_shift), src_cb + (1 << pixel_shift), xchg);
1705 XCHG(top_border + (24 << pixel_shift), src_cb + (9 << pixel_shift), 1);
1706 XCHG(top_border + (32 << pixel_shift), src_cr + (1 << pixel_shift), xchg);
1707 XCHG(top_border + (40 << pixel_shift), src_cr + (9 << pixel_shift), 1);
1708 if (s->mb_x + 1 < s->mb_width) {
1709 XCHG(h->top_borders[top_idx][s->mb_x + 1] + (16 << pixel_shift), src_cb + (17 << pixel_shift), 1);
1710 XCHG(h->top_borders[top_idx][s->mb_x + 1] + (32 << pixel_shift), src_cr + (17 << pixel_shift), 1);
1714 if (deblock_topleft) {
1715 XCHG(top_border_m1 + (16 << pixel_shift), src_cb - (7 << pixel_shift), 1);
1716 XCHG(top_border_m1 + (24 << pixel_shift), src_cr - (7 << pixel_shift), 1);
1718 XCHG(top_border + (16 << pixel_shift), src_cb + 1 + pixel_shift, 1);
1719 XCHG(top_border + (24 << pixel_shift), src_cr + 1 + pixel_shift, 1);
1725 static av_always_inline int dctcoef_get(DCTELEM *mb, int high_bit_depth,
1728 if (high_bit_depth) {
1729 return AV_RN32A(((int32_t *)mb) + index);
1731 return AV_RN16A(mb + index);
1734 static av_always_inline void dctcoef_set(DCTELEM *mb, int high_bit_depth,
1735 int index, int value)
1737 if (high_bit_depth) {
1738 AV_WN32A(((int32_t *)mb) + index, value);
1740 AV_WN16A(mb + index, value);
1743 static av_always_inline void hl_decode_mb_predict_luma(H264Context *h,
1744 int mb_type, int is_h264,
1746 int transform_bypass,
1750 uint8_t *dest_y, int p)
1752 MpegEncContext *const s = &h->s;
1753 void (*idct_add)(uint8_t *dst, DCTELEM *block, int stride);
1754 void (*idct_dc_add)(uint8_t *dst, DCTELEM *block, int stride);
1756 int qscale = p == 0 ? s->qscale : h->chroma_qp[p - 1];
1757 block_offset += 16 * p;
1758 if (IS_INTRA4x4(mb_type)) {
1759 if (simple || !s->encoding) {
1760 if (IS_8x8DCT(mb_type)) {
1761 if (transform_bypass) {
1763 idct_add = s->dsp.add_pixels8;
1765 idct_dc_add = h->h264dsp.h264_idct8_dc_add;
1766 idct_add = h->h264dsp.h264_idct8_add;
1768 for (i = 0; i < 16; i += 4) {
1769 uint8_t *const ptr = dest_y + block_offset[i];
1770 const int dir = h->intra4x4_pred_mode_cache[scan8[i]];
1771 if (transform_bypass && h->sps.profile_idc == 244 && dir <= 1) {
1772 h->hpc.pred8x8l_add[dir](ptr, h->mb + (i * 16 + p * 256 << pixel_shift), linesize);
1774 const int nnz = h->non_zero_count_cache[scan8[i + p * 16]];
1775 h->hpc.pred8x8l[dir](ptr, (h->topleft_samples_available << i) & 0x8000,
1776 (h->topright_samples_available << i) & 0x4000, linesize);
1778 if (nnz == 1 && dctcoef_get(h->mb, pixel_shift, i * 16 + p * 256))
1779 idct_dc_add(ptr, h->mb + (i * 16 + p * 256 << pixel_shift), linesize);
1781 idct_add(ptr, h->mb + (i * 16 + p * 256 << pixel_shift), linesize);
1786 if (transform_bypass) {
1788 idct_add = s->dsp.add_pixels4;
1790 idct_dc_add = h->h264dsp.h264_idct_dc_add;
1791 idct_add = h->h264dsp.h264_idct_add;
1793 for (i = 0; i < 16; i++) {
1794 uint8_t *const ptr = dest_y + block_offset[i];
1795 const int dir = h->intra4x4_pred_mode_cache[scan8[i]];
1797 if (transform_bypass && h->sps.profile_idc == 244 && dir <= 1) {
1798 h->hpc.pred4x4_add[dir](ptr, h->mb + (i * 16 + p * 256 << pixel_shift), linesize);
1803 if (dir == DIAG_DOWN_LEFT_PRED || dir == VERT_LEFT_PRED) {
1804 const int topright_avail = (h->topright_samples_available << i) & 0x8000;
1805 av_assert2(s->mb_y || linesize <= block_offset[i]);
1806 if (!topright_avail) {
1808 tr_high = ((uint16_t *)ptr)[3 - linesize / 2] * 0x0001000100010001ULL;
1809 topright = (uint8_t *)&tr_high;
1811 tr = ptr[3 - linesize] * 0x01010101u;
1812 topright = (uint8_t *)&tr;
1815 topright = ptr + (4 << pixel_shift) - linesize;
1819 h->hpc.pred4x4[dir](ptr, topright, linesize);
1820 nnz = h->non_zero_count_cache[scan8[i + p * 16]];
1823 if (nnz == 1 && dctcoef_get(h->mb, pixel_shift, i * 16 + p * 256))
1824 idct_dc_add(ptr, h->mb + (i * 16 + p * 256 << pixel_shift), linesize);
1826 idct_add(ptr, h->mb + (i * 16 + p * 256 << pixel_shift), linesize);
1827 } else if (CONFIG_SVQ3_DECODER)
1828 ff_svq3_add_idct_c(ptr, h->mb + i * 16 + p * 256, linesize, qscale, 0);
1835 h->hpc.pred16x16[h->intra16x16_pred_mode](dest_y, linesize);
1837 if (h->non_zero_count_cache[scan8[LUMA_DC_BLOCK_INDEX + p]]) {
1838 if (!transform_bypass)
1839 h->h264dsp.h264_luma_dc_dequant_idct(h->mb + (p * 256 << pixel_shift),
1841 h->dequant4_coeff[p][qscale][0]);
1843 static const uint8_t dc_mapping[16] = {
1844 0 * 16, 1 * 16, 4 * 16, 5 * 16,
1845 2 * 16, 3 * 16, 6 * 16, 7 * 16,
1846 8 * 16, 9 * 16, 12 * 16, 13 * 16,
1847 10 * 16, 11 * 16, 14 * 16, 15 * 16 };
1848 for (i = 0; i < 16; i++)
1849 dctcoef_set(h->mb + (p * 256 << pixel_shift),
1850 pixel_shift, dc_mapping[i],
1851 dctcoef_get(h->mb_luma_dc[p],
1855 } else if (CONFIG_SVQ3_DECODER)
1856 ff_svq3_luma_dc_dequant_idct_c(h->mb + p * 256,
1857 h->mb_luma_dc[p], qscale);
1861 static av_always_inline void hl_decode_mb_idct_luma(H264Context *h, int mb_type,
1862 int is_h264, int simple,
1863 int transform_bypass,
1867 uint8_t *dest_y, int p)
1869 MpegEncContext *const s = &h->s;
1870 void (*idct_add)(uint8_t *dst, DCTELEM *block, int stride);
1872 block_offset += 16 * p;
1873 if (!IS_INTRA4x4(mb_type)) {
1875 if (IS_INTRA16x16(mb_type)) {
1876 if (transform_bypass) {
1877 if (h->sps.profile_idc == 244 &&
1878 (h->intra16x16_pred_mode == VERT_PRED8x8 ||
1879 h->intra16x16_pred_mode == HOR_PRED8x8)) {
1880 h->hpc.pred16x16_add[h->intra16x16_pred_mode](dest_y, block_offset,
1881 h->mb + (p * 256 << pixel_shift),
1884 for (i = 0; i < 16; i++)
1885 if (h->non_zero_count_cache[scan8[i + p * 16]] ||
1886 dctcoef_get(h->mb, pixel_shift, i * 16 + p * 256))
1887 s->dsp.add_pixels4(dest_y + block_offset[i],
1888 h->mb + (i * 16 + p * 256 << pixel_shift),
1892 h->h264dsp.h264_idct_add16intra(dest_y, block_offset,
1893 h->mb + (p * 256 << pixel_shift),
1895 h->non_zero_count_cache + p * 5 * 8);
1897 } else if (h->cbp & 15) {
1898 if (transform_bypass) {
1899 const int di = IS_8x8DCT(mb_type) ? 4 : 1;
1900 idct_add = IS_8x8DCT(mb_type) ? s->dsp.add_pixels8
1901 : s->dsp.add_pixels4;
1902 for (i = 0; i < 16; i += di)
1903 if (h->non_zero_count_cache[scan8[i + p * 16]])
1904 idct_add(dest_y + block_offset[i],
1905 h->mb + (i * 16 + p * 256 << pixel_shift),
1908 if (IS_8x8DCT(mb_type))
1909 h->h264dsp.h264_idct8_add4(dest_y, block_offset,
1910 h->mb + (p * 256 << pixel_shift),
1912 h->non_zero_count_cache + p * 5 * 8);
1914 h->h264dsp.h264_idct_add16(dest_y, block_offset,
1915 h->mb + (p * 256 << pixel_shift),
1917 h->non_zero_count_cache + p * 5 * 8);
1920 } else if (CONFIG_SVQ3_DECODER) {
1921 for (i = 0; i < 16; i++)
1922 if (h->non_zero_count_cache[scan8[i + p * 16]] || h->mb[i * 16 + p * 256]) {
1923 // FIXME benchmark weird rule, & below
1924 uint8_t *const ptr = dest_y + block_offset[i];
1925 ff_svq3_add_idct_c(ptr, h->mb + i * 16 + p * 256, linesize,
1926 s->qscale, IS_INTRA(mb_type) ? 1 : 0);
1934 #include "h264_mb_template.c"
1938 #include "h264_mb_template.c"
1942 #include "h264_mb_template.c"
1944 void ff_h264_hl_decode_mb(H264Context *h)
1946 MpegEncContext *const s = &h->s;
1947 const int mb_xy = h->mb_xy;
1948 const int mb_type = s->current_picture.f.mb_type[mb_xy];
1949 int is_complex = CONFIG_SMALL || h->is_complex || IS_INTRA_PCM(mb_type) || s->qscale == 0;
1952 if (is_complex || h->pixel_shift)
1953 hl_decode_mb_444_complex(h);
1955 hl_decode_mb_444_simple_8(h);
1956 } else if (is_complex) {
1957 hl_decode_mb_complex(h);
1958 } else if (h->pixel_shift) {
1959 hl_decode_mb_simple_16(h);
1961 hl_decode_mb_simple_8(h);
1964 static int pred_weight_table(H264Context *h)
1966 MpegEncContext *const s = &h->s;
1968 int luma_def, chroma_def;
1971 h->use_weight_chroma = 0;
1972 h->luma_log2_weight_denom = get_ue_golomb(&s->gb);
1973 if (h->sps.chroma_format_idc)
1974 h->chroma_log2_weight_denom = get_ue_golomb(&s->gb);
1975 luma_def = 1 << h->luma_log2_weight_denom;
1976 chroma_def = 1 << h->chroma_log2_weight_denom;
1978 for (list = 0; list < 2; list++) {
1979 h->luma_weight_flag[list] = 0;
1980 h->chroma_weight_flag[list] = 0;
1981 for (i = 0; i < h->ref_count[list]; i++) {
1982 int luma_weight_flag, chroma_weight_flag;
1984 luma_weight_flag = get_bits1(&s->gb);
1985 if (luma_weight_flag) {
1986 h->luma_weight[i][list][0] = get_se_golomb(&s->gb);
1987 h->luma_weight[i][list][1] = get_se_golomb(&s->gb);
1988 if (h->luma_weight[i][list][0] != luma_def ||
1989 h->luma_weight[i][list][1] != 0) {
1991 h->luma_weight_flag[list] = 1;
1994 h->luma_weight[i][list][0] = luma_def;
1995 h->luma_weight[i][list][1] = 0;
1998 if (h->sps.chroma_format_idc) {
1999 chroma_weight_flag = get_bits1(&s->gb);
2000 if (chroma_weight_flag) {
2002 for (j = 0; j < 2; j++) {
2003 h->chroma_weight[i][list][j][0] = get_se_golomb(&s->gb);
2004 h->chroma_weight[i][list][j][1] = get_se_golomb(&s->gb);
2005 if (h->chroma_weight[i][list][j][0] != chroma_def ||
2006 h->chroma_weight[i][list][j][1] != 0) {
2007 h->use_weight_chroma = 1;
2008 h->chroma_weight_flag[list] = 1;
2013 for (j = 0; j < 2; j++) {
2014 h->chroma_weight[i][list][j][0] = chroma_def;
2015 h->chroma_weight[i][list][j][1] = 0;
2020 if (h->slice_type_nos != AV_PICTURE_TYPE_B)
2023 h->use_weight = h->use_weight || h->use_weight_chroma;
2028 * Initialize implicit_weight table.
2029 * @param field 0/1 initialize the weight for interlaced MBAFF
2030 * -1 initializes the rest
2032 static void implicit_weight_table(H264Context *h, int field)
2034 MpegEncContext *const s = &h->s;
2035 int ref0, ref1, i, cur_poc, ref_start, ref_count0, ref_count1;
2037 for (i = 0; i < 2; i++) {
2038 h->luma_weight_flag[i] = 0;
2039 h->chroma_weight_flag[i] = 0;
2043 if (s->picture_structure == PICT_FRAME) {
2044 cur_poc = s->current_picture_ptr->poc;
2046 cur_poc = s->current_picture_ptr->field_poc[s->picture_structure - 1];
2048 if (h->ref_count[0] == 1 && h->ref_count[1] == 1 && !FRAME_MBAFF &&
2049 h->ref_list[0][0].poc + h->ref_list[1][0].poc == 2 * cur_poc) {
2051 h->use_weight_chroma = 0;
2055 ref_count0 = h->ref_count[0];
2056 ref_count1 = h->ref_count[1];
2058 cur_poc = s->current_picture_ptr->field_poc[field];
2060 ref_count0 = 16 + 2 * h->ref_count[0];
2061 ref_count1 = 16 + 2 * h->ref_count[1];
2065 h->use_weight_chroma = 2;
2066 h->luma_log2_weight_denom = 5;
2067 h->chroma_log2_weight_denom = 5;
2069 for (ref0 = ref_start; ref0 < ref_count0; ref0++) {
2070 int poc0 = h->ref_list[0][ref0].poc;
2071 for (ref1 = ref_start; ref1 < ref_count1; ref1++) {
2073 if (!h->ref_list[0][ref0].long_ref && !h->ref_list[1][ref1].long_ref) {
2074 int poc1 = h->ref_list[1][ref1].poc;
2075 int td = av_clip(poc1 - poc0, -128, 127);
2077 int tb = av_clip(cur_poc - poc0, -128, 127);
2078 int tx = (16384 + (FFABS(td) >> 1)) / td;
2079 int dist_scale_factor = (tb * tx + 32) >> 8;
2080 if (dist_scale_factor >= -64 && dist_scale_factor <= 128)
2081 w = 64 - dist_scale_factor;
2085 h->implicit_weight[ref0][ref1][0] =
2086 h->implicit_weight[ref0][ref1][1] = w;
2088 h->implicit_weight[ref0][ref1][field] = w;
2095 * instantaneous decoder refresh.
2097 static void idr(H264Context *h)
2100 ff_h264_remove_all_refs(h);
2101 h->prev_frame_num = 0;
2102 h->prev_frame_num_offset = 0;
2103 h->prev_poc_msb = 1<<16;
2104 h->prev_poc_lsb = 0;
2105 for (i = 0; i < MAX_DELAYED_PIC_COUNT; i++)
2106 h->last_pocs[i] = INT_MIN;
2109 /* forget old pics after a seek */
2110 static void flush_dpb(AVCodecContext *avctx)
2112 H264Context *h = avctx->priv_data;
2114 for (i=0; i<=MAX_DELAYED_PIC_COUNT; i++) {
2115 if (h->delayed_pic[i])
2116 h->delayed_pic[i]->f.reference = 0;
2117 h->delayed_pic[i] = NULL;
2119 h->outputed_poc = h->next_outputed_poc = INT_MIN;
2120 h->prev_interlaced_frame = 1;
2122 h->prev_frame_num = -1;
2123 if (h->s.current_picture_ptr)
2124 h->s.current_picture_ptr->f.reference = 0;
2125 h->s.first_field = 0;
2126 ff_h264_reset_sei(h);
2127 ff_mpeg_flush(avctx);
2128 h->recovery_frame= -1;
2132 static int init_poc(H264Context *h)
2134 MpegEncContext *const s = &h->s;
2135 const int max_frame_num = 1 << h->sps.log2_max_frame_num;
2137 Picture *cur = s->current_picture_ptr;
2139 h->frame_num_offset = h->prev_frame_num_offset;
2140 if (h->frame_num < h->prev_frame_num)
2141 h->frame_num_offset += max_frame_num;
2143 if (h->sps.poc_type == 0) {
2144 const int max_poc_lsb = 1 << h->sps.log2_max_poc_lsb;
2146 if (h->poc_lsb < h->prev_poc_lsb && h->prev_poc_lsb - h->poc_lsb >= max_poc_lsb / 2)
2147 h->poc_msb = h->prev_poc_msb + max_poc_lsb;
2148 else if (h->poc_lsb > h->prev_poc_lsb && h->prev_poc_lsb - h->poc_lsb < -max_poc_lsb / 2)
2149 h->poc_msb = h->prev_poc_msb - max_poc_lsb;
2151 h->poc_msb = h->prev_poc_msb;
2153 field_poc[1] = h->poc_msb + h->poc_lsb;
2154 if (s->picture_structure == PICT_FRAME)
2155 field_poc[1] += h->delta_poc_bottom;
2156 } else if (h->sps.poc_type == 1) {
2157 int abs_frame_num, expected_delta_per_poc_cycle, expectedpoc;
2160 if (h->sps.poc_cycle_length != 0)
2161 abs_frame_num = h->frame_num_offset + h->frame_num;
2165 if (h->nal_ref_idc == 0 && abs_frame_num > 0)
2168 expected_delta_per_poc_cycle = 0;
2169 for (i = 0; i < h->sps.poc_cycle_length; i++)
2170 // FIXME integrate during sps parse
2171 expected_delta_per_poc_cycle += h->sps.offset_for_ref_frame[i];
2173 if (abs_frame_num > 0) {
2174 int poc_cycle_cnt = (abs_frame_num - 1) / h->sps.poc_cycle_length;
2175 int frame_num_in_poc_cycle = (abs_frame_num - 1) % h->sps.poc_cycle_length;
2177 expectedpoc = poc_cycle_cnt * expected_delta_per_poc_cycle;
2178 for (i = 0; i <= frame_num_in_poc_cycle; i++)
2179 expectedpoc = expectedpoc + h->sps.offset_for_ref_frame[i];
2183 if (h->nal_ref_idc == 0)
2184 expectedpoc = expectedpoc + h->sps.offset_for_non_ref_pic;
2186 field_poc[0] = expectedpoc + h->delta_poc[0];
2187 field_poc[1] = field_poc[0] + h->sps.offset_for_top_to_bottom_field;
2189 if (s->picture_structure == PICT_FRAME)
2190 field_poc[1] += h->delta_poc[1];
2192 int poc = 2 * (h->frame_num_offset + h->frame_num);
2194 if (!h->nal_ref_idc)
2201 if (s->picture_structure != PICT_BOTTOM_FIELD)
2202 s->current_picture_ptr->field_poc[0] = field_poc[0];
2203 if (s->picture_structure != PICT_TOP_FIELD)
2204 s->current_picture_ptr->field_poc[1] = field_poc[1];
2205 cur->poc = FFMIN(cur->field_poc[0], cur->field_poc[1]);
2211 * initialize scan tables
2213 static void init_scan_tables(H264Context *h)
2216 for (i = 0; i < 16; i++) {
2217 #define T(x) (x >> 2) | ((x << 2) & 0xF)
2218 h->zigzag_scan[i] = T(zigzag_scan[i]);
2219 h->field_scan[i] = T(field_scan[i]);
2222 for (i = 0; i < 64; i++) {
2223 #define T(x) (x >> 3) | ((x & 7) << 3)
2224 h->zigzag_scan8x8[i] = T(ff_zigzag_direct[i]);
2225 h->zigzag_scan8x8_cavlc[i] = T(zigzag_scan8x8_cavlc[i]);
2226 h->field_scan8x8[i] = T(field_scan8x8[i]);
2227 h->field_scan8x8_cavlc[i] = T(field_scan8x8_cavlc[i]);
2230 if (h->sps.transform_bypass) { // FIXME same ugly
2231 memcpy(h->zigzag_scan_q0 , zigzag_scan , sizeof(h->zigzag_scan_q0 ));
2232 memcpy(h->zigzag_scan8x8_q0 , ff_zigzag_direct , sizeof(h->zigzag_scan8x8_q0 ));
2233 memcpy(h->zigzag_scan8x8_cavlc_q0 , zigzag_scan8x8_cavlc , sizeof(h->zigzag_scan8x8_cavlc_q0));
2234 memcpy(h->field_scan_q0 , field_scan , sizeof(h->field_scan_q0 ));
2235 memcpy(h->field_scan8x8_q0 , field_scan8x8 , sizeof(h->field_scan8x8_q0 ));
2236 memcpy(h->field_scan8x8_cavlc_q0 , field_scan8x8_cavlc , sizeof(h->field_scan8x8_cavlc_q0 ));
2238 memcpy(h->zigzag_scan_q0 , h->zigzag_scan , sizeof(h->zigzag_scan_q0 ));
2239 memcpy(h->zigzag_scan8x8_q0 , h->zigzag_scan8x8 , sizeof(h->zigzag_scan8x8_q0 ));
2240 memcpy(h->zigzag_scan8x8_cavlc_q0 , h->zigzag_scan8x8_cavlc , sizeof(h->zigzag_scan8x8_cavlc_q0));
2241 memcpy(h->field_scan_q0 , h->field_scan , sizeof(h->field_scan_q0 ));
2242 memcpy(h->field_scan8x8_q0 , h->field_scan8x8 , sizeof(h->field_scan8x8_q0 ));
2243 memcpy(h->field_scan8x8_cavlc_q0 , h->field_scan8x8_cavlc , sizeof(h->field_scan8x8_cavlc_q0 ));
2247 static int field_end(H264Context *h, int in_setup)
2249 MpegEncContext *const s = &h->s;
2250 AVCodecContext *const avctx = s->avctx;
2254 if (!in_setup && !s->dropable)
2255 ff_thread_report_progress(&s->current_picture_ptr->f, INT_MAX,
2256 s->picture_structure == PICT_BOTTOM_FIELD);
2258 if (CONFIG_H264_VDPAU_DECODER &&
2259 s->avctx->codec->capabilities & CODEC_CAP_HWACCEL_VDPAU)
2260 ff_vdpau_h264_set_reference_frames(s);
2262 if (in_setup || !(avctx->active_thread_type & FF_THREAD_FRAME)) {
2264 err = ff_h264_execute_ref_pic_marking(h, h->mmco, h->mmco_index);
2265 h->prev_poc_msb = h->poc_msb;
2266 h->prev_poc_lsb = h->poc_lsb;
2268 h->prev_frame_num_offset = h->frame_num_offset;
2269 h->prev_frame_num = h->frame_num;
2270 h->outputed_poc = h->next_outputed_poc;
2273 if (avctx->hwaccel) {
2274 if (avctx->hwaccel->end_frame(avctx) < 0)
2275 av_log(avctx, AV_LOG_ERROR,
2276 "hardware accelerator failed to decode picture\n");
2279 if (CONFIG_H264_VDPAU_DECODER &&
2280 s->avctx->codec->capabilities & CODEC_CAP_HWACCEL_VDPAU)
2281 ff_vdpau_h264_picture_complete(s);
2284 * FIXME: Error handling code does not seem to support interlaced
2285 * when slices span multiple rows
2286 * The ff_er_add_slice calls don't work right for bottom
2287 * fields; they cause massive erroneous error concealing
2288 * Error marking covers both fields (top and bottom).
2289 * This causes a mismatched s->error_count
2290 * and a bad error table. Further, the error count goes to
2291 * INT_MAX when called for bottom field, because mb_y is
2292 * past end by one (callers fault) and resync_mb_y != 0
2293 * causes problems for the first MB line, too.
2298 ff_MPV_frame_end(s);
2300 h->current_slice = 0;
2306 * Replicate H264 "master" context to thread contexts.
2308 static void clone_slice(H264Context *dst, H264Context *src)
2310 memcpy(dst->block_offset, src->block_offset, sizeof(dst->block_offset));
2311 dst->s.current_picture_ptr = src->s.current_picture_ptr;
2312 dst->s.current_picture = src->s.current_picture;
2313 dst->s.linesize = src->s.linesize;
2314 dst->s.uvlinesize = src->s.uvlinesize;
2315 dst->s.first_field = src->s.first_field;
2317 dst->prev_poc_msb = src->prev_poc_msb;
2318 dst->prev_poc_lsb = src->prev_poc_lsb;
2319 dst->prev_frame_num_offset = src->prev_frame_num_offset;
2320 dst->prev_frame_num = src->prev_frame_num;
2321 dst->short_ref_count = src->short_ref_count;
2323 memcpy(dst->short_ref, src->short_ref, sizeof(dst->short_ref));
2324 memcpy(dst->long_ref, src->long_ref, sizeof(dst->long_ref));
2325 memcpy(dst->default_ref_list, src->default_ref_list, sizeof(dst->default_ref_list));
2326 memcpy(dst->ref_list, src->ref_list, sizeof(dst->ref_list));
2328 memcpy(dst->dequant4_coeff, src->dequant4_coeff, sizeof(src->dequant4_coeff));
2329 memcpy(dst->dequant8_coeff, src->dequant8_coeff, sizeof(src->dequant8_coeff));
2333 * Compute profile from profile_idc and constraint_set?_flags.
2337 * @return profile as defined by FF_PROFILE_H264_*
2339 int ff_h264_get_profile(SPS *sps)
2341 int profile = sps->profile_idc;
2343 switch (sps->profile_idc) {
2344 case FF_PROFILE_H264_BASELINE:
2345 // constraint_set1_flag set to 1
2346 profile |= (sps->constraint_set_flags & 1 << 1) ? FF_PROFILE_H264_CONSTRAINED : 0;
2348 case FF_PROFILE_H264_HIGH_10:
2349 case FF_PROFILE_H264_HIGH_422:
2350 case FF_PROFILE_H264_HIGH_444_PREDICTIVE:
2351 // constraint_set3_flag set to 1
2352 profile |= (sps->constraint_set_flags & 1 << 3) ? FF_PROFILE_H264_INTRA : 0;
2360 * Decode a slice header.
2361 * This will also call ff_MPV_common_init() and frame_start() as needed.
2363 * @param h h264context
2364 * @param h0 h264 master context (differs from 'h' when doing sliced based
2365 * parallel decoding)
2367 * @return 0 if okay, <0 if an error occurred, 1 if decoding must not be multithreaded
2369 static int decode_slice_header(H264Context *h, H264Context *h0)
2371 MpegEncContext *const s = &h->s;
2372 MpegEncContext *const s0 = &h0->s;
2373 unsigned int first_mb_in_slice;
2374 unsigned int pps_id;
2375 int num_ref_idx_active_override_flag;
2376 unsigned int slice_type, tmp, i, j;
2377 int default_ref_list_done = 0;
2378 int last_pic_structure, last_pic_dropable;
2381 /* FIXME: 2tap qpel isn't implemented for high bit depth. */
2382 if ((s->avctx->flags2 & CODEC_FLAG2_FAST) &&
2383 !h->nal_ref_idc && !h->pixel_shift) {
2384 s->me.qpel_put = s->dsp.put_2tap_qpel_pixels_tab;
2385 s->me.qpel_avg = s->dsp.avg_2tap_qpel_pixels_tab;
2387 s->me.qpel_put = s->dsp.put_h264_qpel_pixels_tab;
2388 s->me.qpel_avg = s->dsp.avg_h264_qpel_pixels_tab;
2391 first_mb_in_slice = get_ue_golomb_long(&s->gb);
2393 if (first_mb_in_slice == 0) { // FIXME better field boundary detection
2394 if (h0->current_slice && FIELD_PICTURE) {
2398 h0->current_slice = 0;
2399 if (!s0->first_field) {
2400 if (s->current_picture_ptr && !s->dropable &&
2401 s->current_picture_ptr->owner2 == s) {
2402 ff_thread_report_progress(&s->current_picture_ptr->f, INT_MAX,
2403 s->picture_structure == PICT_BOTTOM_FIELD);
2405 s->current_picture_ptr = NULL;
2409 slice_type = get_ue_golomb_31(&s->gb);
2410 if (slice_type > 9) {
2411 av_log(h->s.avctx, AV_LOG_ERROR,
2412 "slice type too large (%d) at %d %d\n",
2413 slice_type, s->mb_x, s->mb_y);
2416 if (slice_type > 4) {
2418 h->slice_type_fixed = 1;
2420 h->slice_type_fixed = 0;
2422 slice_type = golomb_to_pict_type[slice_type];
2423 if (slice_type == AV_PICTURE_TYPE_I ||
2424 (h0->current_slice != 0 && slice_type == h0->last_slice_type)) {
2425 default_ref_list_done = 1;
2427 h->slice_type = slice_type;
2428 h->slice_type_nos = slice_type & 3;
2430 // to make a few old functions happy, it's wrong though
2431 s->pict_type = h->slice_type;
2433 pps_id = get_ue_golomb(&s->gb);
2434 if (pps_id >= MAX_PPS_COUNT) {
2435 av_log(h->s.avctx, AV_LOG_ERROR, "pps_id %d out of range\n", pps_id);
2438 if (!h0->pps_buffers[pps_id]) {
2439 av_log(h->s.avctx, AV_LOG_ERROR,
2440 "non-existing PPS %u referenced\n",
2444 h->pps = *h0->pps_buffers[pps_id];
2446 if (!h0->sps_buffers[h->pps.sps_id]) {
2447 av_log(h->s.avctx, AV_LOG_ERROR,
2448 "non-existing SPS %u referenced\n",
2452 h->sps = *h0->sps_buffers[h->pps.sps_id];
2454 s->avctx->profile = ff_h264_get_profile(&h->sps);
2455 s->avctx->level = h->sps.level_idc;
2456 s->avctx->refs = h->sps.ref_frame_count;
2458 must_reinit = (s->context_initialized &&
2459 ( 16*h->sps.mb_width != s->avctx->coded_width
2460 || 16*h->sps.mb_height * (2 - h->sps.frame_mbs_only_flag) != s->avctx->coded_height
2461 || s->avctx->bits_per_raw_sample != h->sps.bit_depth_luma
2462 || h->cur_chroma_format_idc != h->sps.chroma_format_idc
2463 || av_cmp_q(h->sps.sar, s->avctx->sample_aspect_ratio)));
2465 if(must_reinit && (h != h0 || (s->avctx->active_thread_type & FF_THREAD_FRAME))) {
2466 av_log_missing_feature(s->avctx,
2467 "Width/height/bit depth/chroma idc changing with threads", 0);
2468 return AVERROR_PATCHWELCOME; // width / height changed during parallelized decoding
2471 s->mb_width = h->sps.mb_width;
2472 s->mb_height = h->sps.mb_height * (2 - h->sps.frame_mbs_only_flag);
2474 h->b_stride = s->mb_width * 4;
2476 s->chroma_y_shift = h->sps.chroma_format_idc <= 1; // 400 uses yuv420p
2478 s->width = 16 * s->mb_width;
2479 s->height = 16 * s->mb_height;
2483 flush_dpb(s->avctx);
2484 ff_MPV_common_end(s);
2486 h->current_slice = 0;
2488 if (!s->context_initialized) {
2490 av_log(h->s.avctx, AV_LOG_ERROR,
2491 "Cannot (re-)initialize context during parallel decoding.\n");
2494 if( FFALIGN(s->avctx->width , 16 ) == s->width
2495 && FFALIGN(s->avctx->height, 16*(2 - h->sps.frame_mbs_only_flag)) == s->height
2496 && !h->sps.crop_right && !h->sps.crop_bottom
2497 && (s->avctx->width != s->width || s->avctx->height && s->height)
2499 av_log(h->s.avctx, AV_LOG_DEBUG, "Using externally provided dimensions\n");
2500 s->avctx->coded_width = s->width;
2501 s->avctx->coded_height = s->height;
2503 avcodec_set_dimensions(s->avctx, s->width, s->height);
2504 s->avctx->width -= (2>>CHROMA444)*FFMIN(h->sps.crop_right, (8<<CHROMA444)-1);
2505 s->avctx->height -= (1<<s->chroma_y_shift)*FFMIN(h->sps.crop_bottom, (16>>s->chroma_y_shift)-1) * (2 - h->sps.frame_mbs_only_flag);
2507 s->avctx->sample_aspect_ratio = h->sps.sar;
2508 av_assert0(s->avctx->sample_aspect_ratio.den);
2510 if (s->avctx->codec->capabilities & CODEC_CAP_HWACCEL_VDPAU
2511 && (h->sps.bit_depth_luma != 8 ||
2512 h->sps.chroma_format_idc > 1)) {
2513 av_log(s->avctx, AV_LOG_ERROR,
2514 "VDPAU decoding does not support video "
2519 if (s->avctx->bits_per_raw_sample != h->sps.bit_depth_luma ||
2520 h->cur_chroma_format_idc != h->sps.chroma_format_idc) {
2521 if (h->sps.bit_depth_luma >= 8 && h->sps.bit_depth_luma <= 14 && h->sps.bit_depth_luma != 11 && h->sps.bit_depth_luma != 13 &&
2522 (h->sps.bit_depth_luma != 9 || !CHROMA422)) {
2523 s->avctx->bits_per_raw_sample = h->sps.bit_depth_luma;
2524 h->cur_chroma_format_idc = h->sps.chroma_format_idc;
2525 h->pixel_shift = h->sps.bit_depth_luma > 8;
2527 ff_h264dsp_init(&h->h264dsp, h->sps.bit_depth_luma, h->sps.chroma_format_idc);
2528 ff_h264_pred_init(&h->hpc, s->codec_id, h->sps.bit_depth_luma, h->sps.chroma_format_idc);
2529 s->dsp.dct_bits = h->sps.bit_depth_luma > 8 ? 32 : 16;
2530 ff_dsputil_init(&s->dsp, s->avctx);
2532 av_log(s->avctx, AV_LOG_ERROR, "Unsupported bit depth: %d chroma_idc: %d\n",
2533 h->sps.bit_depth_luma, h->sps.chroma_format_idc);
2538 if (h->sps.video_signal_type_present_flag) {
2539 s->avctx->color_range = h->sps.full_range>0 ? AVCOL_RANGE_JPEG
2541 if (h->sps.colour_description_present_flag) {
2542 s->avctx->color_primaries = h->sps.color_primaries;
2543 s->avctx->color_trc = h->sps.color_trc;
2544 s->avctx->colorspace = h->sps.colorspace;
2548 if (h->sps.timing_info_present_flag) {
2549 int64_t den = h->sps.time_scale;
2550 if (h->x264_build < 44U)
2552 av_reduce(&s->avctx->time_base.num, &s->avctx->time_base.den,
2553 h->sps.num_units_in_tick, den, 1 << 30);
2556 switch (h->sps.bit_depth_luma) {
2559 if (s->avctx->colorspace == AVCOL_SPC_RGB) {
2560 s->avctx->pix_fmt = AV_PIX_FMT_GBRP9;
2562 s->avctx->pix_fmt = AV_PIX_FMT_YUV444P9;
2563 } else if (CHROMA422)
2564 s->avctx->pix_fmt = AV_PIX_FMT_YUV422P9;
2566 s->avctx->pix_fmt = AV_PIX_FMT_YUV420P9;
2570 if (s->avctx->colorspace == AVCOL_SPC_RGB) {
2571 s->avctx->pix_fmt = AV_PIX_FMT_GBRP10;
2573 s->avctx->pix_fmt = AV_PIX_FMT_YUV444P10;
2574 } else if (CHROMA422)
2575 s->avctx->pix_fmt = AV_PIX_FMT_YUV422P10;
2577 s->avctx->pix_fmt = AV_PIX_FMT_YUV420P10;
2581 if (s->avctx->colorspace == AVCOL_SPC_RGB) {
2582 s->avctx->pix_fmt = AV_PIX_FMT_GBRP12;
2584 s->avctx->pix_fmt = AV_PIX_FMT_YUV444P12;
2585 } else if (CHROMA422)
2586 s->avctx->pix_fmt = AV_PIX_FMT_YUV422P12;
2588 s->avctx->pix_fmt = AV_PIX_FMT_YUV420P12;
2592 if (s->avctx->colorspace == AVCOL_SPC_RGB) {
2593 s->avctx->pix_fmt = AV_PIX_FMT_GBRP14;
2595 s->avctx->pix_fmt = AV_PIX_FMT_YUV444P14;
2596 } else if (CHROMA422)
2597 s->avctx->pix_fmt = AV_PIX_FMT_YUV422P14;
2599 s->avctx->pix_fmt = AV_PIX_FMT_YUV420P14;
2603 s->avctx->pix_fmt = s->avctx->color_range == AVCOL_RANGE_JPEG ? AV_PIX_FMT_YUVJ444P
2604 : AV_PIX_FMT_YUV444P;
2605 if (s->avctx->colorspace == AVCOL_SPC_RGB) {
2606 s->avctx->pix_fmt = AV_PIX_FMT_GBR24P;
2607 av_log(h->s.avctx, AV_LOG_DEBUG, "Detected GBR colorspace.\n");
2608 } else if (s->avctx->colorspace == AVCOL_SPC_YCGCO) {
2609 av_log(h->s.avctx, AV_LOG_WARNING, "Detected unsupported YCgCo colorspace.\n");
2611 } else if (CHROMA422) {
2612 s->avctx->pix_fmt = s->avctx->color_range == AVCOL_RANGE_JPEG ? AV_PIX_FMT_YUVJ422P
2613 : AV_PIX_FMT_YUV422P;
2615 s->avctx->pix_fmt = s->avctx->get_format(s->avctx,
2616 s->avctx->codec->pix_fmts ?
2617 s->avctx->codec->pix_fmts :
2618 s->avctx->color_range == AVCOL_RANGE_JPEG ?
2619 hwaccel_pixfmt_list_h264_jpeg_420 :
2620 ff_hwaccel_pixfmt_list_420);
2624 av_log(s->avctx, AV_LOG_ERROR,
2625 "Unsupported bit depth: %d\n", h->sps.bit_depth_luma);
2626 return AVERROR_INVALIDDATA;
2629 s->avctx->hwaccel = ff_find_hwaccel(s->avctx->codec->id,
2632 if (ff_MPV_common_init(s) < 0) {
2633 av_log(h->s.avctx, AV_LOG_ERROR, "ff_MPV_common_init() failed.\n");
2637 h->prev_interlaced_frame = 1;
2639 init_scan_tables(h);
2640 if (ff_h264_alloc_tables(h) < 0) {
2641 av_log(h->s.avctx, AV_LOG_ERROR,
2642 "Could not allocate memory for h264\n");
2643 return AVERROR(ENOMEM);
2646 if (!HAVE_THREADS || !(s->avctx->active_thread_type & FF_THREAD_SLICE)) {
2647 if (context_init(h) < 0) {
2648 av_log(h->s.avctx, AV_LOG_ERROR, "context_init() failed.\n");
2652 for (i = 1; i < s->slice_context_count; i++) {
2654 c = h->thread_context[i] = av_malloc(sizeof(H264Context));
2655 memcpy(c, h->s.thread_context[i], sizeof(MpegEncContext));
2656 memset(&c->s + 1, 0, sizeof(H264Context) - sizeof(MpegEncContext));
2657 c->h264dsp = h->h264dsp;
2660 c->pixel_shift = h->pixel_shift;
2661 c->cur_chroma_format_idc = h->cur_chroma_format_idc;
2662 init_scan_tables(c);
2663 clone_tables(c, h, i);
2666 for (i = 0; i < s->slice_context_count; i++)
2667 if (context_init(h->thread_context[i]) < 0) {
2668 av_log(h->s.avctx, AV_LOG_ERROR,
2669 "context_init() failed.\n");
2675 if (h == h0 && h->dequant_coeff_pps != pps_id) {
2676 h->dequant_coeff_pps = pps_id;
2677 init_dequant_tables(h);
2680 h->frame_num = get_bits(&s->gb, h->sps.log2_max_frame_num);
2683 h->mb_aff_frame = 0;
2684 last_pic_structure = s0->picture_structure;
2685 last_pic_dropable = s0->dropable;
2686 s->dropable = h->nal_ref_idc == 0;
2687 if (h->sps.frame_mbs_only_flag) {
2688 s->picture_structure = PICT_FRAME;
2690 if (!h->sps.direct_8x8_inference_flag && slice_type == AV_PICTURE_TYPE_B) {
2691 av_log(h->s.avctx, AV_LOG_ERROR, "This stream was generated by a broken encoder, invalid 8x8 inference\n");
2694 if (get_bits1(&s->gb)) { // field_pic_flag
2695 s->picture_structure = PICT_TOP_FIELD + get_bits1(&s->gb); // bottom_field_flag
2697 s->picture_structure = PICT_FRAME;
2698 h->mb_aff_frame = h->sps.mb_aff;
2701 h->mb_field_decoding_flag = s->picture_structure != PICT_FRAME;
2703 if (h0->current_slice != 0) {
2704 if (last_pic_structure != s->picture_structure ||
2705 last_pic_dropable != s->dropable) {
2706 av_log(h->s.avctx, AV_LOG_ERROR,
2707 "Changing field mode (%d -> %d) between slices is not allowed\n",
2708 last_pic_structure, s->picture_structure);
2709 s->picture_structure = last_pic_structure;
2710 s->dropable = last_pic_dropable;
2711 return AVERROR_INVALIDDATA;
2712 } else if (!s0->current_picture_ptr) {
2713 av_log(s->avctx, AV_LOG_ERROR,
2714 "unset current_picture_ptr on %d. slice\n",
2715 h0->current_slice + 1);
2716 return AVERROR_INVALIDDATA;
2719 /* Shorten frame num gaps so we don't have to allocate reference
2720 * frames just to throw them away */
2721 if (h->frame_num != h->prev_frame_num && h->prev_frame_num >= 0) {
2722 int unwrap_prev_frame_num = h->prev_frame_num;
2723 int max_frame_num = 1 << h->sps.log2_max_frame_num;
2725 if (unwrap_prev_frame_num > h->frame_num)
2726 unwrap_prev_frame_num -= max_frame_num;
2728 if ((h->frame_num - unwrap_prev_frame_num) > h->sps.ref_frame_count) {
2729 unwrap_prev_frame_num = (h->frame_num - h->sps.ref_frame_count) - 1;
2730 if (unwrap_prev_frame_num < 0)
2731 unwrap_prev_frame_num += max_frame_num;
2733 h->prev_frame_num = unwrap_prev_frame_num;
2737 /* See if we have a decoded first field looking for a pair...
2738 * Here, we're using that to see if we should mark previously
2739 * decode frames as "finished".
2740 * We have to do that before the "dummy" in-between frame allocation,
2741 * since that can modify s->current_picture_ptr. */
2742 if (s0->first_field) {
2743 assert(s0->current_picture_ptr);
2744 assert(s0->current_picture_ptr->f.data[0]);
2745 assert(s0->current_picture_ptr->f.reference != DELAYED_PIC_REF);
2747 /* Mark old field/frame as completed */
2748 if (!last_pic_dropable && s0->current_picture_ptr->owner2 == s0) {
2749 ff_thread_report_progress(&s0->current_picture_ptr->f, INT_MAX,
2750 last_pic_structure == PICT_BOTTOM_FIELD);
2753 /* figure out if we have a complementary field pair */
2754 if (!FIELD_PICTURE || s->picture_structure == last_pic_structure) {
2755 /* Previous field is unmatched. Don't display it, but let it
2756 * remain for reference if marked as such. */
2757 if (!last_pic_dropable && last_pic_structure != PICT_FRAME) {
2758 ff_thread_report_progress(&s0->current_picture_ptr->f, INT_MAX,
2759 last_pic_structure == PICT_TOP_FIELD);
2762 if (s0->current_picture_ptr->frame_num != h->frame_num) {
2763 /* This and previous field were reference, but had
2764 * different frame_nums. Consider this field first in
2765 * pair. Throw away previous field except for reference
2767 if (!last_pic_dropable && last_pic_structure != PICT_FRAME) {
2768 ff_thread_report_progress(&s0->current_picture_ptr->f, INT_MAX,
2769 last_pic_structure == PICT_TOP_FIELD);
2772 /* Second field in complementary pair */
2773 if (!((last_pic_structure == PICT_TOP_FIELD &&
2774 s->picture_structure == PICT_BOTTOM_FIELD) ||
2775 (last_pic_structure == PICT_BOTTOM_FIELD &&
2776 s->picture_structure == PICT_TOP_FIELD))) {
2777 av_log(s->avctx, AV_LOG_ERROR,
2778 "Invalid field mode combination %d/%d\n",
2779 last_pic_structure, s->picture_structure);
2780 s->picture_structure = last_pic_structure;
2781 s->dropable = last_pic_dropable;
2782 return AVERROR_INVALIDDATA;
2783 } else if (last_pic_dropable != s->dropable) {
2784 av_log(s->avctx, AV_LOG_ERROR,
2785 "Cannot combine reference and non-reference fields in the same frame\n");
2786 av_log_ask_for_sample(s->avctx, NULL);
2787 s->picture_structure = last_pic_structure;
2788 s->dropable = last_pic_dropable;
2789 return AVERROR_INVALIDDATA;
2792 /* Take ownership of this buffer. Note that if another thread owned
2793 * the first field of this buffer, we're not operating on that pointer,
2794 * so the original thread is still responsible for reporting progress
2795 * on that first field (or if that was us, we just did that above).
2796 * By taking ownership, we assign responsibility to ourselves to
2797 * report progress on the second field. */
2798 s0->current_picture_ptr->owner2 = s0;
2803 while (h->frame_num != h->prev_frame_num && h->prev_frame_num >= 0 &&
2804 h->frame_num != (h->prev_frame_num + 1) % (1 << h->sps.log2_max_frame_num)) {
2805 Picture *prev = h->short_ref_count ? h->short_ref[0] : NULL;
2806 av_log(h->s.avctx, AV_LOG_DEBUG, "Frame num gap %d %d\n",
2807 h->frame_num, h->prev_frame_num);
2808 if (ff_h264_frame_start(h) < 0)
2810 h->prev_frame_num++;
2811 h->prev_frame_num %= 1 << h->sps.log2_max_frame_num;
2812 s->current_picture_ptr->frame_num = h->prev_frame_num;
2813 ff_thread_report_progress(&s->current_picture_ptr->f, INT_MAX, 0);
2814 ff_thread_report_progress(&s->current_picture_ptr->f, INT_MAX, 1);
2815 ff_generate_sliding_window_mmcos(h);
2816 if (ff_h264_execute_ref_pic_marking(h, h->mmco, h->mmco_index) < 0 &&
2817 (s->avctx->err_recognition & AV_EF_EXPLODE))
2818 return AVERROR_INVALIDDATA;
2819 /* Error concealment: if a ref is missing, copy the previous ref in its place.
2820 * FIXME: avoiding a memcpy would be nice, but ref handling makes many assumptions
2821 * about there being no actual duplicates.
2822 * FIXME: this doesn't copy padding for out-of-frame motion vectors. Given we're
2823 * concealing a lost frame, this probably isn't noticeable by comparison, but it should
2825 if (h->short_ref_count) {
2827 av_image_copy(h->short_ref[0]->f.data, h->short_ref[0]->f.linesize,
2828 (const uint8_t **)prev->f.data, prev->f.linesize,
2829 s->avctx->pix_fmt, s->mb_width * 16, s->mb_height * 16);
2830 h->short_ref[0]->poc = prev->poc + 2;
2832 h->short_ref[0]->frame_num = h->prev_frame_num;
2836 /* See if we have a decoded first field looking for a pair...
2837 * We're using that to see whether to continue decoding in that
2838 * frame, or to allocate a new one. */
2839 if (s0->first_field) {
2840 assert(s0->current_picture_ptr);
2841 assert(s0->current_picture_ptr->f.data[0]);
2842 assert(s0->current_picture_ptr->f.reference != DELAYED_PIC_REF);
2844 /* figure out if we have a complementary field pair */
2845 if (!FIELD_PICTURE || s->picture_structure == last_pic_structure) {
2846 /* Previous field is unmatched. Don't display it, but let it
2847 * remain for reference if marked as such. */
2848 s0->current_picture_ptr = NULL;
2849 s0->first_field = FIELD_PICTURE;
2851 if (s0->current_picture_ptr->frame_num != h->frame_num) {
2852 ff_thread_report_progress((AVFrame*)s0->current_picture_ptr, INT_MAX,
2853 s0->picture_structure==PICT_BOTTOM_FIELD);
2854 /* This and the previous field had different frame_nums.
2855 * Consider this field first in pair. Throw away previous
2856 * one except for reference purposes. */
2857 s0->first_field = 1;
2858 s0->current_picture_ptr = NULL;
2860 /* Second field in complementary pair */
2861 s0->first_field = 0;
2865 /* Frame or first field in a potentially complementary pair */
2866 s0->first_field = FIELD_PICTURE;
2869 if (!FIELD_PICTURE || s0->first_field) {
2870 if (ff_h264_frame_start(h) < 0) {
2871 s0->first_field = 0;
2875 ff_release_unused_pictures(s, 0);
2881 s->current_picture_ptr->frame_num = h->frame_num; // FIXME frame_num cleanup
2883 av_assert1(s->mb_num == s->mb_width * s->mb_height);
2884 if (first_mb_in_slice << FIELD_OR_MBAFF_PICTURE >= s->mb_num ||
2885 first_mb_in_slice >= s->mb_num) {
2886 av_log(h->s.avctx, AV_LOG_ERROR, "first_mb_in_slice overflow\n");
2889 s->resync_mb_x = s->mb_x = first_mb_in_slice % s->mb_width;
2890 s->resync_mb_y = s->mb_y = (first_mb_in_slice / s->mb_width) << FIELD_OR_MBAFF_PICTURE;
2891 if (s->picture_structure == PICT_BOTTOM_FIELD)
2892 s->resync_mb_y = s->mb_y = s->mb_y + 1;
2893 av_assert1(s->mb_y < s->mb_height);
2895 if (s->picture_structure == PICT_FRAME) {
2896 h->curr_pic_num = h->frame_num;
2897 h->max_pic_num = 1 << h->sps.log2_max_frame_num;
2899 h->curr_pic_num = 2 * h->frame_num + 1;
2900 h->max_pic_num = 1 << (h->sps.log2_max_frame_num + 1);
2903 if (h->nal_unit_type == NAL_IDR_SLICE)
2904 get_ue_golomb(&s->gb); /* idr_pic_id */
2906 if (h->sps.poc_type == 0) {
2907 h->poc_lsb = get_bits(&s->gb, h->sps.log2_max_poc_lsb);
2909 if (h->pps.pic_order_present == 1 && s->picture_structure == PICT_FRAME)
2910 h->delta_poc_bottom = get_se_golomb(&s->gb);
2913 if (h->sps.poc_type == 1 && !h->sps.delta_pic_order_always_zero_flag) {
2914 h->delta_poc[0] = get_se_golomb(&s->gb);
2916 if (h->pps.pic_order_present == 1 && s->picture_structure == PICT_FRAME)
2917 h->delta_poc[1] = get_se_golomb(&s->gb);
2922 if (h->pps.redundant_pic_cnt_present)
2923 h->redundant_pic_count = get_ue_golomb(&s->gb);
2925 // set defaults, might be overridden a few lines later
2926 h->ref_count[0] = h->pps.ref_count[0];
2927 h->ref_count[1] = h->pps.ref_count[1];
2929 if (h->slice_type_nos != AV_PICTURE_TYPE_I) {
2931 max[0] = max[1] = s->picture_structure == PICT_FRAME ? 15 : 31;
2933 if (h->slice_type_nos == AV_PICTURE_TYPE_B)
2934 h->direct_spatial_mv_pred = get_bits1(&s->gb);
2935 num_ref_idx_active_override_flag = get_bits1(&s->gb);
2937 if (num_ref_idx_active_override_flag) {
2938 h->ref_count[0] = get_ue_golomb(&s->gb) + 1;
2939 if (h->slice_type_nos == AV_PICTURE_TYPE_B) {
2940 h->ref_count[1] = get_ue_golomb(&s->gb) + 1;
2942 // full range is spec-ok in this case, even for frames
2943 h->ref_count[1] = 1;
2946 if (h->ref_count[0]-1 > max[0] || h->ref_count[1]-1 > max[1]){
2947 av_log(h->s.avctx, AV_LOG_ERROR, "reference overflow %u > %u or %u > %u\n", h->ref_count[0]-1, max[0], h->ref_count[1]-1, max[1]);
2948 h->ref_count[0] = h->ref_count[1] = 1;
2949 return AVERROR_INVALIDDATA;
2952 if (h->slice_type_nos == AV_PICTURE_TYPE_B)
2957 h->ref_count[1]= h->ref_count[0]= h->list_count= 0;
2959 if (!default_ref_list_done)
2960 ff_h264_fill_default_ref_list(h);
2962 if (h->slice_type_nos != AV_PICTURE_TYPE_I &&
2963 ff_h264_decode_ref_pic_list_reordering(h) < 0) {
2964 h->ref_count[1] = h->ref_count[0] = 0;
2968 if (h->slice_type_nos != AV_PICTURE_TYPE_I) {
2969 s->last_picture_ptr = &h->ref_list[0][0];
2970 s->last_picture_ptr->owner2 = s;
2971 ff_copy_picture(&s->last_picture, s->last_picture_ptr);
2973 if (h->slice_type_nos == AV_PICTURE_TYPE_B) {
2974 s->next_picture_ptr = &h->ref_list[1][0];
2975 s->next_picture_ptr->owner2 = s;
2976 ff_copy_picture(&s->next_picture, s->next_picture_ptr);
2979 if ((h->pps.weighted_pred && h->slice_type_nos == AV_PICTURE_TYPE_P) ||
2980 (h->pps.weighted_bipred_idc == 1 &&
2981 h->slice_type_nos == AV_PICTURE_TYPE_B))
2982 pred_weight_table(h);
2983 else if (h->pps.weighted_bipred_idc == 2 &&
2984 h->slice_type_nos == AV_PICTURE_TYPE_B) {
2985 implicit_weight_table(h, -1);
2988 for (i = 0; i < 2; i++) {
2989 h->luma_weight_flag[i] = 0;
2990 h->chroma_weight_flag[i] = 0;
2994 if (h->nal_ref_idc && ff_h264_decode_ref_pic_marking(h0, &s->gb) < 0 &&
2995 (s->avctx->err_recognition & AV_EF_EXPLODE))
2996 return AVERROR_INVALIDDATA;
2999 ff_h264_fill_mbaff_ref_list(h);
3001 if (h->pps.weighted_bipred_idc == 2 && h->slice_type_nos == AV_PICTURE_TYPE_B) {
3002 implicit_weight_table(h, 0);
3003 implicit_weight_table(h, 1);
3007 if (h->slice_type_nos == AV_PICTURE_TYPE_B && !h->direct_spatial_mv_pred)
3008 ff_h264_direct_dist_scale_factor(h);
3009 ff_h264_direct_ref_list_init(h);
3011 if (h->slice_type_nos != AV_PICTURE_TYPE_I && h->pps.cabac) {
3012 tmp = get_ue_golomb_31(&s->gb);
3014 av_log(s->avctx, AV_LOG_ERROR, "cabac_init_idc overflow\n");
3017 h->cabac_init_idc = tmp;
3020 h->last_qscale_diff = 0;
3021 tmp = h->pps.init_qp + get_se_golomb(&s->gb);
3022 if (tmp > 51 + 6 * (h->sps.bit_depth_luma - 8)) {
3023 av_log(s->avctx, AV_LOG_ERROR, "QP %u out of range\n", tmp);
3027 h->chroma_qp[0] = get_chroma_qp(h, 0, s->qscale);
3028 h->chroma_qp[1] = get_chroma_qp(h, 1, s->qscale);
3029 // FIXME qscale / qp ... stuff
3030 if (h->slice_type == AV_PICTURE_TYPE_SP)
3031 get_bits1(&s->gb); /* sp_for_switch_flag */
3032 if (h->slice_type == AV_PICTURE_TYPE_SP ||
3033 h->slice_type == AV_PICTURE_TYPE_SI)
3034 get_se_golomb(&s->gb); /* slice_qs_delta */
3036 h->deblocking_filter = 1;
3037 h->slice_alpha_c0_offset = 52;
3038 h->slice_beta_offset = 52;
3039 if (h->pps.deblocking_filter_parameters_present) {
3040 tmp = get_ue_golomb_31(&s->gb);
3042 av_log(s->avctx, AV_LOG_ERROR,
3043 "deblocking_filter_idc %u out of range\n", tmp);
3046 h->deblocking_filter = tmp;
3047 if (h->deblocking_filter < 2)
3048 h->deblocking_filter ^= 1; // 1<->0
3050 if (h->deblocking_filter) {
3051 h->slice_alpha_c0_offset += get_se_golomb(&s->gb) << 1;
3052 h->slice_beta_offset += get_se_golomb(&s->gb) << 1;
3053 if (h->slice_alpha_c0_offset > 104U ||
3054 h->slice_beta_offset > 104U) {
3055 av_log(s->avctx, AV_LOG_ERROR,
3056 "deblocking filter parameters %d %d out of range\n",
3057 h->slice_alpha_c0_offset, h->slice_beta_offset);
3063 if (s->avctx->skip_loop_filter >= AVDISCARD_ALL ||
3064 (s->avctx->skip_loop_filter >= AVDISCARD_NONKEY &&
3065 h->slice_type_nos != AV_PICTURE_TYPE_I) ||
3066 (s->avctx->skip_loop_filter >= AVDISCARD_BIDIR &&
3067 h->slice_type_nos == AV_PICTURE_TYPE_B) ||
3068 (s->avctx->skip_loop_filter >= AVDISCARD_NONREF &&
3069 h->nal_ref_idc == 0))
3070 h->deblocking_filter = 0;
3072 if (h->deblocking_filter == 1 && h0->max_contexts > 1) {
3073 if (s->avctx->flags2 & CODEC_FLAG2_FAST) {
3074 /* Cheat slightly for speed:
3075 * Do not bother to deblock across slices. */
3076 h->deblocking_filter = 2;
3078 h0->max_contexts = 1;
3079 if (!h0->single_decode_warning) {
3080 av_log(s->avctx, AV_LOG_INFO,
3081 "Cannot parallelize deblocking type 1, decoding such frames in sequential order\n");
3082 h0->single_decode_warning = 1;
3085 av_log(h->s.avctx, AV_LOG_ERROR,
3086 "Deblocking switched inside frame.\n");
3091 h->qp_thresh = 15 + 52 -
3092 FFMIN(h->slice_alpha_c0_offset, h->slice_beta_offset) -
3094 h->pps.chroma_qp_index_offset[0],
3095 h->pps.chroma_qp_index_offset[1]) +
3096 6 * (h->sps.bit_depth_luma - 8);
3098 h0->last_slice_type = slice_type;
3099 h->slice_num = ++h0->current_slice;
3102 h0->slice_row[(h->slice_num-1)&(MAX_SLICES-1)]= s->resync_mb_y;
3103 if ( h0->slice_row[h->slice_num&(MAX_SLICES-1)] + 3 >= s->resync_mb_y
3104 && h0->slice_row[h->slice_num&(MAX_SLICES-1)] <= s->resync_mb_y
3105 && h->slice_num >= MAX_SLICES) {
3106 //in case of ASO this check needs to be updated depending on how we decide to assign slice numbers in this case
3107 av_log(s->avctx, AV_LOG_WARNING, "Possibly too many slices (%d >= %d), increase MAX_SLICES and recompile if there are artifacts\n", h->slice_num, MAX_SLICES);
3110 for (j = 0; j < 2; j++) {
3112 int *ref2frm = h->ref2frm[h->slice_num & (MAX_SLICES - 1)][j];
3113 for (i = 0; i < 16; i++) {
3115 if (h->ref_list[j][i].f.data[0]) {
3117 uint8_t *base = h->ref_list[j][i].f.base[0];
3118 for (k = 0; k < h->short_ref_count; k++)
3119 if (h->short_ref[k]->f.base[0] == base) {
3123 for (k = 0; k < h->long_ref_count; k++)
3124 if (h->long_ref[k] && h->long_ref[k]->f.base[0] == base) {
3125 id_list[i] = h->short_ref_count + k;
3133 for (i = 0; i < 16; i++)
3134 ref2frm[i + 2] = 4 * id_list[i] +
3135 (h->ref_list[j][i].f.reference & 3);
3137 ref2frm[18 + 1] = -1;
3138 for (i = 16; i < 48; i++)
3139 ref2frm[i + 4] = 4 * id_list[(i - 16) >> 1] +
3140 (h->ref_list[j][i].f.reference & 3);
3143 // FIXME: fix draw_edges + PAFF + frame threads
3144 h->emu_edge_width = (s->flags & CODEC_FLAG_EMU_EDGE ||
3145 (!h->sps.frame_mbs_only_flag &&
3146 s->avctx->active_thread_type))
3148 h->emu_edge_height = (FRAME_MBAFF || FIELD_PICTURE) ? 0 : h->emu_edge_width;
3150 if (s->avctx->debug & FF_DEBUG_PICT_INFO) {
3151 av_log(h->s.avctx, AV_LOG_DEBUG,
3152 "slice:%d %s mb:%d %c%s%s pps:%u frame:%d poc:%d/%d ref:%d/%d qp:%d loop:%d:%d:%d weight:%d%s %s\n",
3154 (s->picture_structure == PICT_FRAME ? "F" : s->picture_structure == PICT_TOP_FIELD ? "T" : "B"),
3156 av_get_picture_type_char(h->slice_type),
3157 h->slice_type_fixed ? " fix" : "",
3158 h->nal_unit_type == NAL_IDR_SLICE ? " IDR" : "",
3159 pps_id, h->frame_num,
3160 s->current_picture_ptr->field_poc[0],
3161 s->current_picture_ptr->field_poc[1],
3162 h->ref_count[0], h->ref_count[1],
3164 h->deblocking_filter,
3165 h->slice_alpha_c0_offset / 2 - 26, h->slice_beta_offset / 2 - 26,
3167 h->use_weight == 1 && h->use_weight_chroma ? "c" : "",
3168 h->slice_type == AV_PICTURE_TYPE_B ? (h->direct_spatial_mv_pred ? "SPAT" : "TEMP") : "");
3174 int ff_h264_get_slice_type(const H264Context *h)
3176 switch (h->slice_type) {
3177 case AV_PICTURE_TYPE_P:
3179 case AV_PICTURE_TYPE_B:
3181 case AV_PICTURE_TYPE_I:
3183 case AV_PICTURE_TYPE_SP:
3185 case AV_PICTURE_TYPE_SI:
3192 static av_always_inline void fill_filter_caches_inter(H264Context *h,
3193 MpegEncContext *const s,
3194 int mb_type, int top_xy,
3195 int left_xy[LEFT_MBS],
3197 int left_type[LEFT_MBS],
3198 int mb_xy, int list)
3200 int b_stride = h->b_stride;
3201 int16_t(*mv_dst)[2] = &h->mv_cache[list][scan8[0]];
3202 int8_t *ref_cache = &h->ref_cache[list][scan8[0]];
3203 if (IS_INTER(mb_type) || IS_DIRECT(mb_type)) {
3204 if (USES_LIST(top_type, list)) {
3205 const int b_xy = h->mb2b_xy[top_xy] + 3 * b_stride;
3206 const int b8_xy = 4 * top_xy + 2;
3207 int (*ref2frm)[64] = (void*)(h->ref2frm[h->slice_table[top_xy] & (MAX_SLICES - 1)][0] + (MB_MBAFF ? 20 : 2));
3208 AV_COPY128(mv_dst - 1 * 8, s->current_picture.f.motion_val[list][b_xy + 0]);
3209 ref_cache[0 - 1 * 8] =
3210 ref_cache[1 - 1 * 8] = ref2frm[list][s->current_picture.f.ref_index[list][b8_xy + 0]];
3211 ref_cache[2 - 1 * 8] =
3212 ref_cache[3 - 1 * 8] = ref2frm[list][s->current_picture.f.ref_index[list][b8_xy + 1]];
3214 AV_ZERO128(mv_dst - 1 * 8);
3215 AV_WN32A(&ref_cache[0 - 1 * 8], ((LIST_NOT_USED) & 0xFF) * 0x01010101u);
3218 if (!IS_INTERLACED(mb_type ^ left_type[LTOP])) {
3219 if (USES_LIST(left_type[LTOP], list)) {
3220 const int b_xy = h->mb2b_xy[left_xy[LTOP]] + 3;
3221 const int b8_xy = 4 * left_xy[LTOP] + 1;
3222 int (*ref2frm)[64] =(void*)( h->ref2frm[h->slice_table[left_xy[LTOP]] & (MAX_SLICES - 1)][0] + (MB_MBAFF ? 20 : 2));
3223 AV_COPY32(mv_dst - 1 + 0, s->current_picture.f.motion_val[list][b_xy + b_stride * 0]);
3224 AV_COPY32(mv_dst - 1 + 8, s->current_picture.f.motion_val[list][b_xy + b_stride * 1]);
3225 AV_COPY32(mv_dst - 1 + 16, s->current_picture.f.motion_val[list][b_xy + b_stride * 2]);
3226 AV_COPY32(mv_dst - 1 + 24, s->current_picture.f.motion_val[list][b_xy + b_stride * 3]);
3228 ref_cache[-1 + 8] = ref2frm[list][s->current_picture.f.ref_index[list][b8_xy + 2 * 0]];
3229 ref_cache[-1 + 16] =
3230 ref_cache[-1 + 24] = ref2frm[list][s->current_picture.f.ref_index[list][b8_xy + 2 * 1]];
3232 AV_ZERO32(mv_dst - 1 + 0);
3233 AV_ZERO32(mv_dst - 1 + 8);
3234 AV_ZERO32(mv_dst - 1 + 16);
3235 AV_ZERO32(mv_dst - 1 + 24);
3238 ref_cache[-1 + 16] =
3239 ref_cache[-1 + 24] = LIST_NOT_USED;
3244 if (!USES_LIST(mb_type, list)) {
3245 fill_rectangle(mv_dst, 4, 4, 8, pack16to32(0, 0), 4);
3246 AV_WN32A(&ref_cache[0 * 8], ((LIST_NOT_USED) & 0xFF) * 0x01010101u);
3247 AV_WN32A(&ref_cache[1 * 8], ((LIST_NOT_USED) & 0xFF) * 0x01010101u);
3248 AV_WN32A(&ref_cache[2 * 8], ((LIST_NOT_USED) & 0xFF) * 0x01010101u);
3249 AV_WN32A(&ref_cache[3 * 8], ((LIST_NOT_USED) & 0xFF) * 0x01010101u);
3254 int8_t *ref = &s->current_picture.f.ref_index[list][4 * mb_xy];
3255 int (*ref2frm)[64] = (void*)(h->ref2frm[h->slice_num & (MAX_SLICES - 1)][0] + (MB_MBAFF ? 20 : 2));
3256 uint32_t ref01 = (pack16to32(ref2frm[list][ref[0]], ref2frm[list][ref[1]]) & 0x00FF00FF) * 0x0101;
3257 uint32_t ref23 = (pack16to32(ref2frm[list][ref[2]], ref2frm[list][ref[3]]) & 0x00FF00FF) * 0x0101;
3258 AV_WN32A(&ref_cache[0 * 8], ref01);
3259 AV_WN32A(&ref_cache[1 * 8], ref01);
3260 AV_WN32A(&ref_cache[2 * 8], ref23);
3261 AV_WN32A(&ref_cache[3 * 8], ref23);
3265 int16_t(*mv_src)[2] = &s->current_picture.f.motion_val[list][4 * s->mb_x + 4 * s->mb_y * b_stride];
3266 AV_COPY128(mv_dst + 8 * 0, mv_src + 0 * b_stride);
3267 AV_COPY128(mv_dst + 8 * 1, mv_src + 1 * b_stride);
3268 AV_COPY128(mv_dst + 8 * 2, mv_src + 2 * b_stride);
3269 AV_COPY128(mv_dst + 8 * 3, mv_src + 3 * b_stride);
3275 * @return non zero if the loop filter can be skipped
3277 static int fill_filter_caches(H264Context *h, int mb_type)
3279 MpegEncContext *const s = &h->s;
3280 const int mb_xy = h->mb_xy;
3281 int top_xy, left_xy[LEFT_MBS];
3282 int top_type, left_type[LEFT_MBS];
3286 top_xy = mb_xy - (s->mb_stride << MB_FIELD);
3288 /* Wow, what a mess, why didn't they simplify the interlacing & intra
3289 * stuff, I can't imagine that these complex rules are worth it. */
3291 left_xy[LBOT] = left_xy[LTOP] = mb_xy - 1;
3293 const int left_mb_field_flag = IS_INTERLACED(s->current_picture.f.mb_type[mb_xy - 1]);
3294 const int curr_mb_field_flag = IS_INTERLACED(mb_type);
3296 if (left_mb_field_flag != curr_mb_field_flag)
3297 left_xy[LTOP] -= s->mb_stride;
3299 if (curr_mb_field_flag)
3300 top_xy += s->mb_stride &
3301 (((s->current_picture.f.mb_type[top_xy] >> 7) & 1) - 1);
3302 if (left_mb_field_flag != curr_mb_field_flag)
3303 left_xy[LBOT] += s->mb_stride;
3307 h->top_mb_xy = top_xy;
3308 h->left_mb_xy[LTOP] = left_xy[LTOP];
3309 h->left_mb_xy[LBOT] = left_xy[LBOT];
3311 /* For sufficiently low qp, filtering wouldn't do anything.
3312 * This is a conservative estimate: could also check beta_offset
3313 * and more accurate chroma_qp. */
3314 int qp_thresh = h->qp_thresh; // FIXME strictly we should store qp_thresh for each mb of a slice
3315 int qp = s->current_picture.f.qscale_table[mb_xy];
3316 if (qp <= qp_thresh &&
3317 (left_xy[LTOP] < 0 ||
3318 ((qp + s->current_picture.f.qscale_table[left_xy[LTOP]] + 1) >> 1) <= qp_thresh) &&
3320 ((qp + s->current_picture.f.qscale_table[top_xy] + 1) >> 1) <= qp_thresh)) {
3323 if ((left_xy[LTOP] < 0 ||
3324 ((qp + s->current_picture.f.qscale_table[left_xy[LBOT]] + 1) >> 1) <= qp_thresh) &&
3325 (top_xy < s->mb_stride ||
3326 ((qp + s->current_picture.f.qscale_table[top_xy - s->mb_stride] + 1) >> 1) <= qp_thresh))
3331 top_type = s->current_picture.f.mb_type[top_xy];
3332 left_type[LTOP] = s->current_picture.f.mb_type[left_xy[LTOP]];
3333 left_type[LBOT] = s->current_picture.f.mb_type[left_xy[LBOT]];
3334 if (h->deblocking_filter == 2) {
3335 if (h->slice_table[top_xy] != h->slice_num)
3337 if (h->slice_table[left_xy[LBOT]] != h->slice_num)
3338 left_type[LTOP] = left_type[LBOT] = 0;
3340 if (h->slice_table[top_xy] == 0xFFFF)
3342 if (h->slice_table[left_xy[LBOT]] == 0xFFFF)
3343 left_type[LTOP] = left_type[LBOT] = 0;
3345 h->top_type = top_type;
3346 h->left_type[LTOP] = left_type[LTOP];
3347 h->left_type[LBOT] = left_type[LBOT];
3349 if (IS_INTRA(mb_type))
3352 fill_filter_caches_inter(h, s, mb_type, top_xy, left_xy,
3353 top_type, left_type, mb_xy, 0);
3354 if (h->list_count == 2)
3355 fill_filter_caches_inter(h, s, mb_type, top_xy, left_xy,
3356 top_type, left_type, mb_xy, 1);
3358 nnz = h->non_zero_count[mb_xy];
3359 nnz_cache = h->non_zero_count_cache;
3360 AV_COPY32(&nnz_cache[4 + 8 * 1], &nnz[0]);
3361 AV_COPY32(&nnz_cache[4 + 8 * 2], &nnz[4]);
3362 AV_COPY32(&nnz_cache[4 + 8 * 3], &nnz[8]);
3363 AV_COPY32(&nnz_cache[4 + 8 * 4], &nnz[12]);
3364 h->cbp = h->cbp_table[mb_xy];
3367 nnz = h->non_zero_count[top_xy];
3368 AV_COPY32(&nnz_cache[4 + 8 * 0], &nnz[3 * 4]);
3371 if (left_type[LTOP]) {
3372 nnz = h->non_zero_count[left_xy[LTOP]];
3373 nnz_cache[3 + 8 * 1] = nnz[3 + 0 * 4];
3374 nnz_cache[3 + 8 * 2] = nnz[3 + 1 * 4];
3375 nnz_cache[3 + 8 * 3] = nnz[3 + 2 * 4];
3376 nnz_cache[3 + 8 * 4] = nnz[3 + 3 * 4];
3379 /* CAVLC 8x8dct requires NNZ values for residual decoding that differ
3380 * from what the loop filter needs */
3381 if (!CABAC && h->pps.transform_8x8_mode) {
3382 if (IS_8x8DCT(top_type)) {
3383 nnz_cache[4 + 8 * 0] =
3384 nnz_cache[5 + 8 * 0] = (h->cbp_table[top_xy] & 0x4000) >> 12;
3385 nnz_cache[6 + 8 * 0] =
3386 nnz_cache[7 + 8 * 0] = (h->cbp_table[top_xy] & 0x8000) >> 12;
3388 if (IS_8x8DCT(left_type[LTOP])) {
3389 nnz_cache[3 + 8 * 1] =
3390 nnz_cache[3 + 8 * 2] = (h->cbp_table[left_xy[LTOP]] & 0x2000) >> 12; // FIXME check MBAFF
3392 if (IS_8x8DCT(left_type[LBOT])) {
3393 nnz_cache[3 + 8 * 3] =
3394 nnz_cache[3 + 8 * 4] = (h->cbp_table[left_xy[LBOT]] & 0x8000) >> 12; // FIXME check MBAFF
3397 if (IS_8x8DCT(mb_type)) {
3398 nnz_cache[scan8[0]] =
3399 nnz_cache[scan8[1]] =
3400 nnz_cache[scan8[2]] =
3401 nnz_cache[scan8[3]] = (h->cbp & 0x1000) >> 12;
3403 nnz_cache[scan8[0 + 4]] =
3404 nnz_cache[scan8[1 + 4]] =
3405 nnz_cache[scan8[2 + 4]] =
3406 nnz_cache[scan8[3 + 4]] = (h->cbp & 0x2000) >> 12;
3408 nnz_cache[scan8[0 + 8]] =
3409 nnz_cache[scan8[1 + 8]] =
3410 nnz_cache[scan8[2 + 8]] =
3411 nnz_cache[scan8[3 + 8]] = (h->cbp & 0x4000) >> 12;
3413 nnz_cache[scan8[0 + 12]] =
3414 nnz_cache[scan8[1 + 12]] =
3415 nnz_cache[scan8[2 + 12]] =
3416 nnz_cache[scan8[3 + 12]] = (h->cbp & 0x8000) >> 12;
3423 static void loop_filter(H264Context *h, int start_x, int end_x)
3425 MpegEncContext *const s = &h->s;
3426 uint8_t *dest_y, *dest_cb, *dest_cr;
3427 int linesize, uvlinesize, mb_x, mb_y;
3428 const int end_mb_y = s->mb_y + FRAME_MBAFF;
3429 const int old_slice_type = h->slice_type;
3430 const int pixel_shift = h->pixel_shift;
3431 const int block_h = 16 >> s->chroma_y_shift;
3433 if (h->deblocking_filter) {
3434 for (mb_x = start_x; mb_x < end_x; mb_x++)
3435 for (mb_y = end_mb_y - FRAME_MBAFF; mb_y <= end_mb_y; mb_y++) {
3437 mb_xy = h->mb_xy = mb_x + mb_y * s->mb_stride;
3438 h->slice_num = h->slice_table[mb_xy];
3439 mb_type = s->current_picture.f.mb_type[mb_xy];
3440 h->list_count = h->list_counts[mb_xy];
3444 h->mb_field_decoding_flag = !!IS_INTERLACED(mb_type);
3448 dest_y = s->current_picture.f.data[0] +
3449 ((mb_x << pixel_shift) + mb_y * s->linesize) * 16;
3450 dest_cb = s->current_picture.f.data[1] +
3451 (mb_x << pixel_shift) * (8 << CHROMA444) +
3452 mb_y * s->uvlinesize * block_h;
3453 dest_cr = s->current_picture.f.data[2] +
3454 (mb_x << pixel_shift) * (8 << CHROMA444) +
3455 mb_y * s->uvlinesize * block_h;
3456 // FIXME simplify above
3459 linesize = h->mb_linesize = s->linesize * 2;
3460 uvlinesize = h->mb_uvlinesize = s->uvlinesize * 2;
3461 if (mb_y & 1) { // FIXME move out of this function?
3462 dest_y -= s->linesize * 15;
3463 dest_cb -= s->uvlinesize * (block_h - 1);
3464 dest_cr -= s->uvlinesize * (block_h - 1);
3467 linesize = h->mb_linesize = s->linesize;
3468 uvlinesize = h->mb_uvlinesize = s->uvlinesize;
3470 backup_mb_border(h, dest_y, dest_cb, dest_cr, linesize,
3472 if (fill_filter_caches(h, mb_type))
3474 h->chroma_qp[0] = get_chroma_qp(h, 0, s->current_picture.f.qscale_table[mb_xy]);
3475 h->chroma_qp[1] = get_chroma_qp(h, 1, s->current_picture.f.qscale_table[mb_xy]);
3478 ff_h264_filter_mb(h, mb_x, mb_y, dest_y, dest_cb, dest_cr,
3479 linesize, uvlinesize);
3481 ff_h264_filter_mb_fast(h, mb_x, mb_y, dest_y, dest_cb,
3482 dest_cr, linesize, uvlinesize);
3486 h->slice_type = old_slice_type;
3488 s->mb_y = end_mb_y - FRAME_MBAFF;
3489 h->chroma_qp[0] = get_chroma_qp(h, 0, s->qscale);
3490 h->chroma_qp[1] = get_chroma_qp(h, 1, s->qscale);
3493 static void predict_field_decoding_flag(H264Context *h)
3495 MpegEncContext *const s = &h->s;
3496 const int mb_xy = s->mb_x + s->mb_y * s->mb_stride;
3497 int mb_type = (h->slice_table[mb_xy - 1] == h->slice_num) ?
3498 s->current_picture.f.mb_type[mb_xy - 1] :
3499 (h->slice_table[mb_xy - s->mb_stride] == h->slice_num) ?
3500 s->current_picture.f.mb_type[mb_xy - s->mb_stride] : 0;
3501 h->mb_mbaff = h->mb_field_decoding_flag = IS_INTERLACED(mb_type) ? 1 : 0;
3505 * Draw edges and report progress for the last MB row.
3507 static void decode_finish_row(H264Context *h)
3509 MpegEncContext *const s = &h->s;
3510 int top = 16 * (s->mb_y >> FIELD_PICTURE);
3511 int pic_height = 16 * s->mb_height >> FIELD_PICTURE;
3512 int height = 16 << FRAME_MBAFF;
3513 int deblock_border = (16 + 4) << FRAME_MBAFF;
3515 if (h->deblocking_filter) {
3516 if ((top + height) >= pic_height)
3517 height += deblock_border;
3518 top -= deblock_border;
3521 if (top >= pic_height || (top + height) < h->emu_edge_height)
3524 height = FFMIN(height, pic_height - top);
3525 if (top < h->emu_edge_height) {
3526 height = top + height;
3530 ff_draw_horiz_band(s, top, height);
3535 ff_thread_report_progress(&s->current_picture_ptr->f, top + height - 1,
3536 s->picture_structure == PICT_BOTTOM_FIELD);
3539 static int decode_slice(struct AVCodecContext *avctx, void *arg)
3541 H264Context *h = *(void **)arg;
3542 MpegEncContext *const s = &h->s;
3543 const int part_mask = s->partitioned_frame ? (ER_AC_END | ER_AC_ERROR)
3545 int lf_x_start = s->mb_x;
3547 s->mb_skip_run = -1;
3549 h->is_complex = FRAME_MBAFF || s->picture_structure != PICT_FRAME ||
3550 s->codec_id != AV_CODEC_ID_H264 ||
3551 (CONFIG_GRAY && (s->flags & CODEC_FLAG_GRAY));
3555 align_get_bits(&s->gb);
3558 ff_init_cabac_decoder(&h->cabac,
3559 s->gb.buffer + get_bits_count(&s->gb) / 8,
3560 (get_bits_left(&s->gb) + 7) / 8);
3562 ff_h264_init_cabac_states(h);
3566 int ret = ff_h264_decode_mb_cabac(h);
3568 // STOP_TIMER("decode_mb_cabac")
3571 ff_h264_hl_decode_mb(h);
3573 // FIXME optimal? or let mb_decode decode 16x32 ?
3574 if (ret >= 0 && FRAME_MBAFF) {
3577 ret = ff_h264_decode_mb_cabac(h);
3580 ff_h264_hl_decode_mb(h);
3583 eos = get_cabac_terminate(&h->cabac);
3585 if ((s->workaround_bugs & FF_BUG_TRUNCATED) &&
3586 h->cabac.bytestream > h->cabac.bytestream_end + 2) {
3587 ff_er_add_slice(s, s->resync_mb_x, s->resync_mb_y, s->mb_x - 1,
3588 s->mb_y, ER_MB_END & part_mask);
3589 if (s->mb_x >= lf_x_start)
3590 loop_filter(h, lf_x_start, s->mb_x + 1);
3593 if (h->cabac.bytestream > h->cabac.bytestream_end + 2 )
3594 av_log(h->s.avctx, AV_LOG_DEBUG, "bytestream overread %td\n", h->cabac.bytestream_end - h->cabac.bytestream);
3595 if (ret < 0 || h->cabac.bytestream > h->cabac.bytestream_end + 4) {
3596 av_log(h->s.avctx, AV_LOG_ERROR,
3597 "error while decoding MB %d %d, bytestream (%td)\n",
3599 h->cabac.bytestream_end - h->cabac.bytestream);
3600 ff_er_add_slice(s, s->resync_mb_x, s->resync_mb_y, s->mb_x,
3601 s->mb_y, ER_MB_ERROR & part_mask);
3605 if (++s->mb_x >= s->mb_width) {
3606 loop_filter(h, lf_x_start, s->mb_x);
3607 s->mb_x = lf_x_start = 0;
3608 decode_finish_row(h);
3610 if (FIELD_OR_MBAFF_PICTURE) {
3612 if (FRAME_MBAFF && s->mb_y < s->mb_height)
3613 predict_field_decoding_flag(h);
3617 if (eos || s->mb_y >= s->mb_height) {
3618 tprintf(s->avctx, "slice end %d %d\n",
3619 get_bits_count(&s->gb), s->gb.size_in_bits);
3620 ff_er_add_slice(s, s->resync_mb_x, s->resync_mb_y, s->mb_x - 1,
3621 s->mb_y, ER_MB_END & part_mask);
3622 if (s->mb_x > lf_x_start)
3623 loop_filter(h, lf_x_start, s->mb_x);
3629 int ret = ff_h264_decode_mb_cavlc(h);
3632 ff_h264_hl_decode_mb(h);
3634 // FIXME optimal? or let mb_decode decode 16x32 ?
3635 if (ret >= 0 && FRAME_MBAFF) {
3637 ret = ff_h264_decode_mb_cavlc(h);
3640 ff_h264_hl_decode_mb(h);
3645 av_log(h->s.avctx, AV_LOG_ERROR,
3646 "error while decoding MB %d %d\n", s->mb_x, s->mb_y);
3647 ff_er_add_slice(s, s->resync_mb_x, s->resync_mb_y, s->mb_x,
3648 s->mb_y, ER_MB_ERROR & part_mask);
3652 if (++s->mb_x >= s->mb_width) {
3653 loop_filter(h, lf_x_start, s->mb_x);
3654 s->mb_x = lf_x_start = 0;
3655 decode_finish_row(h);
3657 if (FIELD_OR_MBAFF_PICTURE) {
3659 if (FRAME_MBAFF && s->mb_y < s->mb_height)
3660 predict_field_decoding_flag(h);
3662 if (s->mb_y >= s->mb_height) {
3663 tprintf(s->avctx, "slice end %d %d\n",
3664 get_bits_count(&s->gb), s->gb.size_in_bits);
3666 if ( get_bits_left(&s->gb) == 0
3667 || get_bits_left(&s->gb) > 0 && !(s->avctx->err_recognition & AV_EF_AGGRESSIVE)) {
3668 ff_er_add_slice(s, s->resync_mb_x, s->resync_mb_y,
3669 s->mb_x - 1, s->mb_y,
3670 ER_MB_END & part_mask);
3674 ff_er_add_slice(s, s->resync_mb_x, s->resync_mb_y,
3676 ER_MB_END & part_mask);
3683 if (get_bits_left(&s->gb) <= 0 && s->mb_skip_run <= 0) {
3684 tprintf(s->avctx, "slice end %d %d\n",
3685 get_bits_count(&s->gb), s->gb.size_in_bits);
3686 if (get_bits_left(&s->gb) == 0) {
3687 ff_er_add_slice(s, s->resync_mb_x, s->resync_mb_y,
3688 s->mb_x - 1, s->mb_y,
3689 ER_MB_END & part_mask);
3690 if (s->mb_x > lf_x_start)
3691 loop_filter(h, lf_x_start, s->mb_x);
3695 ff_er_add_slice(s, s->resync_mb_x, s->resync_mb_y, s->mb_x,
3696 s->mb_y, ER_MB_ERROR & part_mask);
3706 * Call decode_slice() for each context.
3708 * @param h h264 master context
3709 * @param context_count number of contexts to execute
3711 static int execute_decode_slices(H264Context *h, int context_count)
3713 MpegEncContext *const s = &h->s;
3714 AVCodecContext *const avctx = s->avctx;
3718 if (s->avctx->hwaccel ||
3719 s->avctx->codec->capabilities & CODEC_CAP_HWACCEL_VDPAU)
3721 if (context_count == 1) {
3722 return decode_slice(avctx, &h);
3724 for (i = 1; i < context_count; i++) {
3725 hx = h->thread_context[i];
3726 hx->s.err_recognition = avctx->err_recognition;
3727 hx->s.error_count = 0;
3728 hx->x264_build = h->x264_build;
3731 avctx->execute(avctx, decode_slice, h->thread_context,
3732 NULL, context_count, sizeof(void *));
3734 /* pull back stuff from slices to master context */
3735 hx = h->thread_context[context_count - 1];
3736 s->mb_x = hx->s.mb_x;
3737 s->mb_y = hx->s.mb_y;
3738 s->dropable = hx->s.dropable;
3739 s->picture_structure = hx->s.picture_structure;
3740 for (i = 1; i < context_count; i++)
3741 h->s.error_count += h->thread_context[i]->s.error_count;
3747 static int decode_nal_units(H264Context *h, const uint8_t *buf, int buf_size)
3749 MpegEncContext *const s = &h->s;
3750 AVCodecContext *const avctx = s->avctx;
3751 H264Context *hx; ///< thread context
3755 int pass = !(avctx->active_thread_type & FF_THREAD_FRAME);
3756 int nals_needed = 0; ///< number of NALs that need decoding before the next frame thread starts
3759 h->nal_unit_type= 0;
3761 if(!s->slice_context_count)
3762 s->slice_context_count= 1;
3763 h->max_contexts = s->slice_context_count;
3764 if (!(s->flags2 & CODEC_FLAG2_CHUNKS)) {
3765 h->current_slice = 0;
3766 if (!s->first_field)
3767 s->current_picture_ptr = NULL;
3768 ff_h264_reset_sei(h);
3771 if (h->nal_length_size == 4) {
3772 if (buf_size > 8 && AV_RB32(buf) == 1 && AV_RB32(buf+5) > (unsigned)buf_size) {
3774 }else if(buf_size > 3 && AV_RB32(buf) > 1 && AV_RB32(buf) <= (unsigned)buf_size)
3778 for (; pass <= 1; pass++) {
3781 next_avc = h->is_avc ? 0 : buf_size;
3791 if (buf_index >= next_avc) {
3792 if (buf_index >= buf_size - h->nal_length_size)
3795 for (i = 0; i < h->nal_length_size; i++)
3796 nalsize = (nalsize << 8) | buf[buf_index++];
3797 if (nalsize <= 0 || nalsize > buf_size - buf_index) {
3798 av_log(h->s.avctx, AV_LOG_ERROR,
3799 "AVC: nal size %d\n", nalsize);
3802 next_avc = buf_index + nalsize;
3804 // start code prefix search
3805 for (; buf_index + 3 < next_avc; buf_index++)
3806 // This should always succeed in the first iteration.
3807 if (buf[buf_index] == 0 &&
3808 buf[buf_index + 1] == 0 &&
3809 buf[buf_index + 2] == 1)
3812 if (buf_index + 3 >= buf_size) {
3813 buf_index = buf_size;
3818 if (buf_index >= next_avc)
3822 hx = h->thread_context[context_count];
3824 ptr = ff_h264_decode_nal(hx, buf + buf_index, &dst_length,
3825 &consumed, next_avc - buf_index);
3826 if (ptr == NULL || dst_length < 0) {
3830 i = buf_index + consumed;
3831 if ((s->workaround_bugs & FF_BUG_AUTODETECT) && i + 3 < next_avc &&
3832 buf[i] == 0x00 && buf[i + 1] == 0x00 &&
3833 buf[i + 2] == 0x01 && buf[i + 3] == 0xE0)
3834 s->workaround_bugs |= FF_BUG_TRUNCATED;
3836 if (!(s->workaround_bugs & FF_BUG_TRUNCATED))
3837 while(dst_length > 0 && ptr[dst_length - 1] == 0)
3839 bit_length = !dst_length ? 0
3841 decode_rbsp_trailing(h, ptr + dst_length - 1));
3843 if (s->avctx->debug & FF_DEBUG_STARTCODE)
3844 av_log(h->s.avctx, AV_LOG_DEBUG, "NAL %d/%d at %d/%d length %d pass %d\n", hx->nal_unit_type, hx->nal_ref_idc, buf_index, buf_size, dst_length, pass);
3846 if (h->is_avc && (nalsize != consumed) && nalsize)
3847 av_log(h->s.avctx, AV_LOG_DEBUG,
3848 "AVC: Consumed only %d bytes instead of %d\n",
3851 buf_index += consumed;
3855 /* packets can sometimes contain multiple PPS/SPS,
3856 * e.g. two PAFF field pictures in one packet, or a demuxer
3857 * which splits NALs strangely if so, when frame threading we
3858 * can't start the next thread until we've read all of them */
3859 switch (hx->nal_unit_type) {
3862 nals_needed = nal_index;
3866 init_get_bits(&hx->s.gb, ptr, bit_length);
3867 if (!get_ue_golomb(&hx->s.gb))
3868 nals_needed = nal_index;
3873 // FIXME do not discard SEI id
3874 if (avctx->skip_frame >= AVDISCARD_NONREF && h->nal_ref_idc == 0)
3880 if (h->decoding_extradata) {
3881 switch (hx->nal_unit_type) {
3887 case NAL_AUXILIARY_SLICE:
3888 av_log(h->s.avctx, AV_LOG_WARNING, "Ignoring NAL %d in global header\n", hx->nal_unit_type);
3889 hx->nal_unit_type = NAL_FILLER_DATA;
3893 switch (hx->nal_unit_type) {
3895 if (h->nal_unit_type != NAL_IDR_SLICE) {
3896 av_log(h->s.avctx, AV_LOG_ERROR,
3897 "Invalid mix of idr and non-idr slices\n");
3901 idr(h); // FIXME ensure we don't lose some frames if there is reordering
3903 init_get_bits(&hx->s.gb, ptr, bit_length);
3905 hx->inter_gb_ptr = &hx->s.gb;
3906 hx->s.data_partitioning = 0;
3908 if ((err = decode_slice_header(hx, h)))
3911 if (h->sei_recovery_frame_cnt >= 0 && (h->frame_num != h->sei_recovery_frame_cnt || hx->slice_type_nos != AV_PICTURE_TYPE_I))
3912 h->valid_recovery_point = 1;
3914 if ( h->sei_recovery_frame_cnt >= 0
3915 && ( h->recovery_frame<0
3916 || ((h->recovery_frame - h->frame_num) & ((1 << h->sps.log2_max_frame_num)-1)) > h->sei_recovery_frame_cnt)) {
3917 h->recovery_frame = (h->frame_num + h->sei_recovery_frame_cnt) %
3918 (1 << h->sps.log2_max_frame_num);
3920 if (!h->valid_recovery_point)
3921 h->recovery_frame = h->frame_num;
3924 s->current_picture_ptr->f.key_frame |=
3925 (hx->nal_unit_type == NAL_IDR_SLICE);
3927 if (h->recovery_frame == h->frame_num) {
3928 s->current_picture_ptr->sync |= 1;
3929 h->recovery_frame = -1;
3932 h->sync |= !!s->current_picture_ptr->f.key_frame;
3933 h->sync |= 3*!!(s->flags2 & CODEC_FLAG2_SHOW_ALL);
3934 s->current_picture_ptr->sync |= h->sync;
3936 if (h->current_slice == 1) {
3937 if (!(s->flags2 & CODEC_FLAG2_CHUNKS))
3938 decode_postinit(h, nal_index >= nals_needed);
3940 if (s->avctx->hwaccel &&
3941 s->avctx->hwaccel->start_frame(s->avctx, NULL, 0) < 0)
3943 if (CONFIG_H264_VDPAU_DECODER &&
3944 s->avctx->codec->capabilities & CODEC_CAP_HWACCEL_VDPAU)
3945 ff_vdpau_h264_picture_start(s);
3948 if (hx->redundant_pic_count == 0 &&
3949 (avctx->skip_frame < AVDISCARD_NONREF ||
3951 (avctx->skip_frame < AVDISCARD_BIDIR ||
3952 hx->slice_type_nos != AV_PICTURE_TYPE_B) &&
3953 (avctx->skip_frame < AVDISCARD_NONKEY ||
3954 hx->slice_type_nos == AV_PICTURE_TYPE_I) &&
3955 avctx->skip_frame < AVDISCARD_ALL) {
3956 if (avctx->hwaccel) {
3957 if (avctx->hwaccel->decode_slice(avctx,
3958 &buf[buf_index - consumed],
3961 } else if (CONFIG_H264_VDPAU_DECODER &&
3962 s->avctx->codec->capabilities & CODEC_CAP_HWACCEL_VDPAU) {
3963 static const uint8_t start_code[] = {
3965 ff_vdpau_add_data_chunk(s, start_code,
3966 sizeof(start_code));
3967 ff_vdpau_add_data_chunk(s, &buf[buf_index - consumed],
3974 init_get_bits(&hx->s.gb, ptr, bit_length);
3976 hx->inter_gb_ptr = NULL;
3978 if ((err = decode_slice_header(hx, h)) < 0)
3981 hx->s.data_partitioning = 1;
3984 init_get_bits(&hx->intra_gb, ptr, bit_length);
3985 hx->intra_gb_ptr = &hx->intra_gb;
3988 init_get_bits(&hx->inter_gb, ptr, bit_length);
3989 hx->inter_gb_ptr = &hx->inter_gb;
3991 av_log(h->s.avctx, AV_LOG_ERROR, "Partitioned H.264 support is incomplete\n");
3994 if (hx->redundant_pic_count == 0 &&
3996 hx->s.data_partitioning &&
3997 s->current_picture_ptr &&
3998 s->context_initialized &&
3999 (avctx->skip_frame < AVDISCARD_NONREF || hx->nal_ref_idc) &&
4000 (avctx->skip_frame < AVDISCARD_BIDIR ||
4001 hx->slice_type_nos != AV_PICTURE_TYPE_B) &&
4002 (avctx->skip_frame < AVDISCARD_NONKEY ||
4003 hx->slice_type_nos == AV_PICTURE_TYPE_I) &&
4004 avctx->skip_frame < AVDISCARD_ALL)
4008 init_get_bits(&s->gb, ptr, bit_length);
4009 ff_h264_decode_sei(h);
4012 init_get_bits(&s->gb, ptr, bit_length);
4013 if (ff_h264_decode_seq_parameter_set(h) < 0 && (h->is_avc ? (nalsize != consumed) && nalsize : 1)) {
4014 av_log(h->s.avctx, AV_LOG_DEBUG,
4015 "SPS decoding failure, trying again with the complete NAL\n");
4017 av_assert0(next_avc - buf_index + consumed == nalsize);
4018 if ((next_avc - buf_index + consumed - 1) >= INT_MAX/8)
4020 init_get_bits(&s->gb, &buf[buf_index + 1 - consumed],
4021 8*(next_avc - buf_index + consumed - 1));
4022 ff_h264_decode_seq_parameter_set(h);
4025 if (s->flags & CODEC_FLAG_LOW_DELAY ||
4026 (h->sps.bitstream_restriction_flag &&
4027 !h->sps.num_reorder_frames)) {
4028 if (s->avctx->has_b_frames > 1 || h->delayed_pic[0])
4029 av_log(avctx, AV_LOG_WARNING, "Delayed frames seen "
4030 "reenabling low delay requires a codec "
4036 if (avctx->has_b_frames < 2)
4037 avctx->has_b_frames = !s->low_delay;
4040 init_get_bits(&s->gb, ptr, bit_length);
4041 ff_h264_decode_picture_parameter_set(h, bit_length);
4044 case NAL_END_SEQUENCE:
4045 case NAL_END_STREAM:
4046 case NAL_FILLER_DATA:
4048 case NAL_AUXILIARY_SLICE:
4051 av_log(avctx, AV_LOG_DEBUG, "Unknown NAL code: %d (%d bits)\n",
4052 hx->nal_unit_type, bit_length);
4055 if (context_count == h->max_contexts) {
4056 execute_decode_slices(h, context_count);
4061 av_log(h->s.avctx, AV_LOG_ERROR, "decode_slice_header error\n");
4062 else if (err == 1) {
4063 /* Slice could not be decoded in parallel mode, copy down
4064 * NAL unit stuff to context 0 and restart. Note that
4065 * rbsp_buffer is not transferred, but since we no longer
4066 * run in parallel mode this should not be an issue. */
4067 h->nal_unit_type = hx->nal_unit_type;
4068 h->nal_ref_idc = hx->nal_ref_idc;
4075 execute_decode_slices(h, context_count);
4079 if (s->current_picture_ptr && s->current_picture_ptr->owner2 == s &&
4081 ff_thread_report_progress(&s->current_picture_ptr->f, INT_MAX,
4082 s->picture_structure == PICT_BOTTOM_FIELD);
4089 * Return the number of bytes consumed for building the current frame.
4091 static int get_consumed_bytes(MpegEncContext *s, int pos, int buf_size)
4094 pos = 1; // avoid infinite loops (i doubt that is needed but ...)
4095 if (pos + 10 > buf_size)
4096 pos = buf_size; // oops ;)
4101 static int decode_frame(AVCodecContext *avctx, void *data,
4102 int *got_frame, AVPacket *avpkt)
4104 const uint8_t *buf = avpkt->data;
4105 int buf_size = avpkt->size;
4106 H264Context *h = avctx->priv_data;
4107 MpegEncContext *s = &h->s;
4108 AVFrame *pict = data;
4113 s->flags = avctx->flags;
4114 s->flags2 = avctx->flags2;
4116 /* end of stream, output what is still in the buffers */
4117 if (buf_size == 0) {
4120 s->current_picture_ptr = NULL;
4122 // FIXME factorize this with the output code below
4123 out = h->delayed_pic[0];
4126 h->delayed_pic[i] &&
4127 !h->delayed_pic[i]->f.key_frame &&
4128 !h->delayed_pic[i]->mmco_reset;
4130 if (h->delayed_pic[i]->poc < out->poc) {
4131 out = h->delayed_pic[i];
4135 for (i = out_idx; h->delayed_pic[i]; i++)
4136 h->delayed_pic[i] = h->delayed_pic[i + 1];
4145 if(h->is_avc && buf_size >= 9 && buf[0]==1 && buf[2]==0 && (buf[4]&0xFC)==0xFC && (buf[5]&0x1F) && buf[8]==0x67){
4146 int cnt= buf[5]&0x1f;
4147 const uint8_t *p= buf+6;
4149 int nalsize= AV_RB16(p) + 2;
4150 if(nalsize > buf_size - (p-buf) || p[2]!=0x67)
4158 int nalsize= AV_RB16(p) + 2;
4159 if(nalsize > buf_size - (p-buf) || p[2]!=0x68)
4164 return ff_h264_decode_extradata(h, buf, buf_size);
4168 buf_index = decode_nal_units(h, buf, buf_size);
4172 if (!s->current_picture_ptr && h->nal_unit_type == NAL_END_SEQUENCE) {
4173 av_assert0(buf_index <= buf_size);
4177 if (!(s->flags2 & CODEC_FLAG2_CHUNKS) && !s->current_picture_ptr) {
4178 if (avctx->skip_frame >= AVDISCARD_NONREF ||
4179 buf_size >= 4 && !memcmp("Q264", buf, 4))
4181 av_log(avctx, AV_LOG_ERROR, "no frame!\n");
4185 if (!(s->flags2 & CODEC_FLAG2_CHUNKS) ||
4186 (s->mb_y >= s->mb_height && s->mb_height)) {
4187 if (s->flags2 & CODEC_FLAG2_CHUNKS)
4188 decode_postinit(h, 1);
4192 /* Wait for second field. */
4194 if (h->next_output_pic && (h->next_output_pic->sync || h->sync>1)) {
4196 *pict = h->next_output_pic->f;
4200 assert(pict->data[0] || !*got_frame);
4201 ff_print_debug_info(s, pict);
4203 return get_consumed_bytes(s, buf_index, buf_size);
4206 av_cold void ff_h264_free_context(H264Context *h)
4210 free_tables(h, 1); // FIXME cleanup init stuff perhaps
4212 for (i = 0; i < MAX_SPS_COUNT; i++)
4213 av_freep(h->sps_buffers + i);
4215 for (i = 0; i < MAX_PPS_COUNT; i++)
4216 av_freep(h->pps_buffers + i);
4219 static av_cold int h264_decode_end(AVCodecContext *avctx)
4221 H264Context *h = avctx->priv_data;
4222 MpegEncContext *s = &h->s;
4224 ff_h264_remove_all_refs(h);
4225 ff_h264_free_context(h);
4227 ff_MPV_common_end(s);
4229 // memset(h, 0, sizeof(H264Context));
4234 static const AVProfile profiles[] = {
4235 { FF_PROFILE_H264_BASELINE, "Baseline" },
4236 { FF_PROFILE_H264_CONSTRAINED_BASELINE, "Constrained Baseline" },
4237 { FF_PROFILE_H264_MAIN, "Main" },
4238 { FF_PROFILE_H264_EXTENDED, "Extended" },
4239 { FF_PROFILE_H264_HIGH, "High" },
4240 { FF_PROFILE_H264_HIGH_10, "High 10" },
4241 { FF_PROFILE_H264_HIGH_10_INTRA, "High 10 Intra" },
4242 { FF_PROFILE_H264_HIGH_422, "High 4:2:2" },
4243 { FF_PROFILE_H264_HIGH_422_INTRA, "High 4:2:2 Intra" },
4244 { FF_PROFILE_H264_HIGH_444, "High 4:4:4" },
4245 { FF_PROFILE_H264_HIGH_444_PREDICTIVE, "High 4:4:4 Predictive" },
4246 { FF_PROFILE_H264_HIGH_444_INTRA, "High 4:4:4 Intra" },
4247 { FF_PROFILE_H264_CAVLC_444, "CAVLC 4:4:4" },
4248 { FF_PROFILE_UNKNOWN },
4251 static const AVOption h264_options[] = {
4252 {"is_avc", "is avc", offsetof(H264Context, is_avc), FF_OPT_TYPE_INT, {.i64 = 0}, 0, 1, 0},
4253 {"nal_length_size", "nal_length_size", offsetof(H264Context, nal_length_size), FF_OPT_TYPE_INT, {.i64 = 0}, 0, 4, 0},
4257 static const AVClass h264_class = {
4258 .class_name = "H264 Decoder",
4259 .item_name = av_default_item_name,
4260 .option = h264_options,
4261 .version = LIBAVUTIL_VERSION_INT,
4264 static const AVClass h264_vdpau_class = {
4265 .class_name = "H264 VDPAU Decoder",
4266 .item_name = av_default_item_name,
4267 .option = h264_options,
4268 .version = LIBAVUTIL_VERSION_INT,
4271 AVCodec ff_h264_decoder = {
4273 .type = AVMEDIA_TYPE_VIDEO,
4274 .id = AV_CODEC_ID_H264,
4275 .priv_data_size = sizeof(H264Context),
4276 .init = ff_h264_decode_init,
4277 .close = h264_decode_end,
4278 .decode = decode_frame,
4279 .capabilities = /*CODEC_CAP_DRAW_HORIZ_BAND |*/ CODEC_CAP_DR1 |
4280 CODEC_CAP_DELAY | CODEC_CAP_SLICE_THREADS |
4281 CODEC_CAP_FRAME_THREADS,
4283 .long_name = NULL_IF_CONFIG_SMALL("H.264 / AVC / MPEG-4 AVC / MPEG-4 part 10"),
4284 .init_thread_copy = ONLY_IF_THREADS_ENABLED(decode_init_thread_copy),
4285 .update_thread_context = ONLY_IF_THREADS_ENABLED(decode_update_thread_context),
4286 .profiles = NULL_IF_CONFIG_SMALL(profiles),
4287 .priv_class = &h264_class,
4290 #if CONFIG_H264_VDPAU_DECODER
4291 AVCodec ff_h264_vdpau_decoder = {
4292 .name = "h264_vdpau",
4293 .type = AVMEDIA_TYPE_VIDEO,
4294 .id = AV_CODEC_ID_H264,
4295 .priv_data_size = sizeof(H264Context),
4296 .init = ff_h264_decode_init,
4297 .close = h264_decode_end,
4298 .decode = decode_frame,
4299 .capabilities = CODEC_CAP_DR1 | CODEC_CAP_DELAY | CODEC_CAP_HWACCEL_VDPAU,
4301 .long_name = NULL_IF_CONFIG_SMALL("H.264 / AVC / MPEG-4 AVC / MPEG-4 part 10 (VDPAU acceleration)"),
4302 .pix_fmts = (const enum AVPixelFormat[]) { AV_PIX_FMT_VDPAU_H264,
4304 .profiles = NULL_IF_CONFIG_SMALL(profiles),
4305 .priv_class = &h264_vdpau_class,